Evidence on the long-term - WHO

Evidence on the long-term
effects of breastfeeding

SYSTEMATIC REVIEWS AND META-ANALYSES

Bernardo L. Horta, MD, PhD
Universidade Federal de Pelotas, Pelotas, Brazil
Rajiv Bahl, MD, PhD
Department of Child and Adolescent Health and Development,
World Health Organization, Geneva, Switzerland
José C. Martines, MD, PhD
Department of Child and Adolescent Health and Development,
World Health Organization, Geneva, Switzerland
Cesar G. Victora, MD, PhD
Universidade Federal de Pelotas, Pelotas, Brazil

i

WHO Library Cataloguing-in-Publication Data

Evidence on the long-term effects of breastfeeding : systematic review and meta-analyses /
Bernardo L. Horta ... [et al.].

1.Breast feeding. 2.Blood pressure. 3.Diabetes mellitus. 4.Cholesterol. 5.Obesity. I.Horta,
Bernardo L. II.World Health Organization.

ISBN 978 92 4 159523 0 (NLM classification: WS 125)

© World Health Organization 2007

All rights reserved. Publications of the World Health Organization can be obtained from WHO
Press, World Health Organization, 20 Avenue Appia, 1211 Geneva 27, Switzerland (tel.:
+41 22 791 3264; fax: +41 22 791 4857; e-mail: [email protected]). Requests for
permission to reproduce or translate WHO publications – whether for sale or for noncommercial
distribution – should be addressed to WHO Press, at the above address (fax: +41 22 791
4806; e-mail: [email protected]).

The designations employed and the presentation of the material in this publication do not
imply the expression of any opinion whatsoever on the part of the World Health Organization
concerning the legal status of any country, territory, city or area or of its authorities, or concerning
the delimitation of its frontiers or boundaries. Dotted lines on maps represent approximate
border lines for which there may not yet be full agreement.

The mention of specific companies or of certain manufacturers’ products does not imply that
they are endorsed or recommended by the World Health Organization in preference to others
of a similar nature that are not mentioned. Errors and omissions excepted, the names of
proprietary products are distinguished by initial capital letters.

All reasonable precautions have been taken by the World Health Organization to verify the
information contained in this publication. However, the published material is being distributed
without warranty of any kind, either expressed or implied. The responsibility for the
interpretation and use of the material lies with the reader. In no event shall the World Health
Organization be liable for damages arising from its use.

The named authors alone are responsible for the views expressed in this publication.
ii EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Contents

Executive summary 1
I. Introduction 3
II. General methodological issues in studies of the long-term effects of breastfeeding 4
4
Study design 4
Factors affecting the internal validity of individual studies 6
Main sources of heterogeneity among studies 8
III. Search methods 8
Selection criteria for studies 8
Types of outcome measures 8
Search strategy 9
IV. Review methods 9
Assessment of study quality 9
Data abstraction 9
Data analysis 11
V. Results and discussion 11
Review 1 – Breastfeeding and blood pressure in later life 20
Review 2 – Breastfeeding and blood cholesterol in later life 25
Review 3 – Breastfeeding and the risk of overweight and obesity in later life 34
Review 4 – Breastfeeding and the risk of type-2 diabetes 36
Review 5 – Breastfeeding and school achievement/intelligence levels 40
VI. Conclusions 42
References

iii

iv EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Executive summary

Background: Breastfeeding presents clear short-term benefits for child health, mainly protection
against morbidity and mortality from infectious diseases. On the other hand, there is some
controversy on the long-term consequences of breastfeeding. Whereas some studies reported that
breastfed subjects present a higher level of school achievement and performance in intelligence
tests, as well as lower blood pressure, lower total cholesterol and a lower prevalence of overweight
and obesity, others have failed to detect such associations.

Objectives: The primary objective of this series of systematic reviews was to assess the effects of
breastfeeding on blood pressure, diabetes and related indicators, serum cholesterol, overweight and
obesity, and intellectual performance.

Search strategy: Two independent literature searches were conducted at the World Health
Organization in Geneva, Switzerland, and at the University of Pelotas in Brazil, comprising the
MEDLINE (1966 to March 2006) and Scientific Citation Index databases.

Selection criteria: We selected observational and randomized studies, published in English, French,
Portuguese and Spanish, assessing the effects of breastfeeding on blood pressure, obesity/overweight,
total cholesterol, type-2 diabetes, and intellectual performance. Studies that restricted the measurement
of outcomes to infancy were excluded from the meta-analyses. The type of comparison group used
(e.g. never breastfed or breastfed for less than x months) did not constitute a selection criterion.

Data extraction and analysis: Two reviewers independently evaluated study quality, using a
standardized protocol, and disagreement was resolved by consensus rating. Fixed and random-effects
models were used to pool the effect estimates, and a random-effects regression was used to assess
several potential sources of heterogeneity.

Effect on blood pressure: We included 30 and 25 estimates for systolic and diastolic blood pressure,
respectively. In a random-effects model, systolic (mean difference: -1.21 mmHg; 95% confidence
interval (CI): -1.72 to -0.70) and diastolic blood pressures (mean difference: -0.49 mm Hg; 95% CI:
-0.87 to -0.11) were lower among breastfed subjects. Publication bias was evident, with smaller
studies reporting a greater protective effect of breastfeeding. However, even among studies with
>1000 participants a statistically significant effect of breastfeeding was observed (mean difference
in systolic blood pressure: -0.59 mmHg; 95% CI: -1.00 to -0.19). Adjustment for confounding was
also a source of heterogeneity between study results, but even among those studies controlling for
several socioeconomic and demographic variables, systolic (mean difference: -1.19; 95% CI: -1.70 to
-0.69) and diastolic (mean difference: -0.61; 95% CI: -1.12 to -0.10) blood pressures were lower
among breastfed subjects. Publication bias and residual confounding may be responsible for part
(but not all) of the observed effect of breastfeeding on blood pressure.

Effect on serum cholesterol: Breastfed subjects presented lower mean total cholesterol in adulthood
(mean difference: -0.18; 95% CI: -0.30 to -0.06 mmol/L), whereas for children and adolescents the
association was not statistically significant. Age at assessment of cholesterol explained about 60% of
the heterogeneity between studies, whereas study size, control for confounding, year of birth and

1

categorization of breastfeeding duration did not play a significant role. The evidence suggests that
breastfeeding is related to lower cholesterol levels and this association is not due to publication bias
or residual confounding.

Effect on overweight and obesity: We obtained 39 estimates of the effect of breastfeeding on
prevalence of overweight/obesity. In a random-effects model, breastfed individuals were less likely to
be considered as overweight and/or obese, with a pooled odds ratio of 0.78 (95% CI: 0.72–0.84).
Control for confounding, age at assessment, year of birth, and study design did not modify the effect
of breastfeeding. Because a statistically significant protective effect was observed among those
studies that controlled for socioeconomic status and parental anthropometry, as well as with >1500
participants, the effect of breastfeeding was not likely to be due to publication bias or confounding.

Effect on type-2 diabetes: We identified five papers that evaluated the relationship between
breastfeeding duration and type-2 diabetes. Breastfed subjects were less likely to present type-2
diabetes (pooled odds ratio: 0.63; 95% CI: 0.45–0.89).

Effect on intelligence and schooling: For the assessment of performance in intelligence tests, we
obtained data from eight studies that controlled for intellectual stimulation at home and collected
information on infant feeding in infancy, in which the duration of breastfeeding was of at least one
month among breastfed subjects. Performance in intelligence tests was higher among those subjects
who had been breastfed (mean difference: 4.9; 95% CI: 2.97–6.92). Positive studies included a
randomized trial. Regarding school performance in late adolescence or young adulthood, three studies
showed a positive effect of breastfeeding.

Limitations: Because nearly all studies included in the analyses are observational, it is not possible
to completely rule out the possibility that these results may be partly explained by self-selection of
breastfeeding mothers or by residual confounding. Publication bias was assessed by examining the
effect of study size on the estimates and was found not to be important for most outcomes. Very
few studies were available from low/middle-income countries, where the effect of breastfeeding may
be modified by social and cultural conditions.

Reviewers’ conclusions: The available evidence suggests that breastfeeding may have long-term
benefits. Subjects who were breastfed experienced lower mean blood pressure and total cholesterol,
as well as higher performance in intelligence tests. Furthermore, the prevalence of overweight/obesity
and type-2 diabetes was lower among breastfed subjects. All effects were statistically significant, but
for some outcomes their magnitude was relatively modest.

2 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

I. Introduction

Breastfeeding brings clear short-term benefits natal exposures – such as infant growth and feed-
for child health by reducing mortality and mor- ing patterns – that could also be related to the
bidity from infectious diseases. A collaborative development of chronic diseases.
reanalysis of studies conducted in middle/low-
income countries reported a reduced risk of The notion that nutrition during early phases
mortality from infectious diseases among of human development can alter organ function,
breastfed infants, up to the second birthday (1). and thereby predispose – or programme – indi-
Kramer et al (2) reviewed the evidence on the viduals to a later onset of adult disease, is an
effects on child health and growth of exclusive area of considerable interest to researchers and
breastfeeding for 6 months. Infants who were of great concern to public health. This idea origi-
exclusively breastfed for 6 months presented nates from the more general concept in devel-
lower morbidity from gastrointestinal and al- opmental biology which was defined by Lucas
lergic diseases, while showing similar growth as “programming” (11). This is defined as the
rates to non-breastfed children. process whereby a stimulus or insult applied at
a critical or sensitive period of development re-
Based on such evidence, WHO (3) and sults in a long-term or permanent effect on the
UNICEF (4) now recommend that every infant structure or function of the organism. This hy-
should be exclusively breastfed for the first six pothesis is currently described as the “develop-
months of life, with continued breastfeeding for mental origins of health and disease” (12).
up to two years or longer. In this review we ad-
dress the long-term consequences of breastfeeding Over 400 scientific publications are avail-
on adult health and intellectual development. able on the association between breastfeeding
and health outcomes beyond infancy. Some re-
Current interest in the long-term conse- searchers claim that the benefits of breastfeeding
quences of early life exposures has been fuelled include increased school achievement or per-
by the original finding of Barker et al (5) that formance in intelligence tests, reduced mean
size at birth and in infancy was related to the blood pressure, lower total cholesterol, and a
development of adult diseases – including dia- lower prevalence of overweight and obesity. On
betes, hypertension and cardiovascular condi- the other hand, other studies have failed to de-
tions. These findings led to the fetal origin hy- tect such associations. The evidence on long-
pothesis, which postulates that adverse intrau- term effects of breastfeeding may be important
terine conditions would be responsible for fetal for further promotion of this healthy practice
malnutrition and low birthweight, a process that throughout the world.
would also increase the susceptibility to chronic
diseases in adulthood. Indeed, epidemiological The Department of Child and Adolescent
studies in several countries have reported in- Health and Development in the World Health
creased risks of chronic diseases (6-8) among Organization commissioned the present system-
adults who were small at birth. atic review of the available evidence on long-
term consequences of breastfeeding. The follow-
Because many studies on the long-term con- ing long-term outcomes of public health impor-
sequences of intrauterine growth may be affected tance were examined: blood pressure, diabetes
by confounding variables – particularly socioeco- and related indicators, serum cholesterol, over-
nomic status – and by inappropriate statistical weight and obesity, and intellectual performance.
analyses, some authors challenge whether or not These outcomes are of great interest to research-
these associations are causal (9). On the other ers, as evidenced by the number of publications
hand, Lucas et al (10) pointed out that meth- identified. This report describes the methods,
odological flaws in studies of intrauterine growth results and conclusions of this review.
may have deflected attention from important post-

3

II. General methodological issues in studies
of the long-term effects of breastfeeding

Study design Search for evidence on the long-term conse-
quences of breastfeeding should not be restricted
The strength of scientific inference depends on to randomized trials, because of their small
the internal validity of the study. Randomized number. It must involve tracking down the best
controlled trials, if properly designed and con- available studies with rigorous design; prospec-
ducted, are considered as the gold standard of tive birth cohort studies should be considered
design validity, being less susceptible than other as the next best design in terms of strength of
designs to selection and information bias, as well evidence. It has been shown that the results of
as to confounding (13). Furthermore, there are meta-analysis of well-designed observational
clearly defined standards for conducting and studies, with either cohort or case-control de-
reporting on randomized clinical trials, all in- sign, can be remarkably similar to that of
tended to increase the validity of their results randomized controlled trials on the same topic
and interpretation (14). (17). Nevertheless, birth cohort studies are sus-
ceptible to self-selection and confounding, is-
As mentioned previously, breastfeeding has sues that will be discussed below.
clear short-term benefits – i.e. it reduces mor-
bidity and mortality from infectious diseases (1) In the next section, the main methodologi-
– and it is now unethical to randomly allocate cal issues affecting these studies will be de-
infants to breastmilk or formula. However, about scribed, as well as the strategies used to mini-
20 years ago the evidence supporting breastfeed- mize the impact of these limitations on the find-
ing was not so clear-cut and randomized trials ings of the present review.
could be carried out. In a British study from
1982, pre-term infants were randomly assigned Factors affecting the internal
to formula or banked breastmilk; recent follow- validity of individual studies
up of these subjects has provided an opportu-
nity to assess whether breastfeeding can pro- In the context of studies on long-term conse-
gramme the later occurrence of risk factors for quences of breastfeeding, the following issues
cardiovascular diseases (15). should be considered.

Most recently, in the Promotion of Losses to follow-up
Breastfeeding Intervention Trial (16) in Belarus,
maternal hospitals and their corresponding poly- A major source of bias in cohort studies as well
clinics were randomly assigned to implement or as randomized controlled trials with long-term
not to implement the Baby-Friendly Hospital outcomes relates to the need for follow-up of
Initiative. Duration and exclusivity of individuals for a period of time after exposure
breastfeeding were higher in the intervention in order to assess the occurrence of the out-
group (16). Because breastfeeding promotion comes of interest. If a large proportion of sub-
was randomized, rather than breastfeeding per jects are lost during follow-up, the study’s valid-
se, the trial was ethically sound. Follow-up of ity is reduced. Baseline data, such as breastfeed-
these children will provide an excellent oppor- ing status, should be examined to determine
tunity for studying the long-term effects of whether there are systematic differences be-
breastfeeding. It should be noted, however, that tween subjects who were followed up and those
the Belarus study has relatively low statistical who were not; if the losses are similar accord-
power because compliance with breastfeeding ing to the baseline characteristics, selection bias
promotion was far from perfect. is unlikely (18).

4 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Misclassification among the poor (22). Thus, confounding by so-
cioeconomic status may underestimate the ben-
When the methods used for obtaining informa- eficial effects of breastfeeding – for example, on
tion on either infant feeding or the outcomes educational attainment – because breastfed sub-
are inaccurate, misclassification may occur. This jects will tend to be poorer. Depending on the
may take two forms: differential and non-dif- association between cardiovascular risk factors
ferential misclassification. and wealth in these societies, confounding can
act in either direction. For example, if high cho-
Misclassification of breastfeeding duration lesterol levels are more frequent among the rich,
is more likely in retrospective than in prospec- the protective effect of breastfeeding will be
tive designs. Huttly et al (19), in a prospective overestimated.
study, compared the actual breastfeeding dura-
tion with the duration reported retrospectively Even if confounding factors are controlled
by the mothers. They observed a systematic bias through multivariable analyses, there is a possi-
towards reporting longer durations of bility of residual confounding. Inaccurate meas-
breastfeeding for wealthier and more educated urement of confounders, as well as incorrect
mothers, while those from low socioeconomic specification of statistical models, may preclude
status families did not tend to err more in one full adjustment for confounding and lead to es-
direction than in the other. Because high socio- timates of the impact of breastfeeding that are
economic status is related to a lower prevalence biased. For example, if information on family
of cardiovascular diseases, such differential income is not precise, control for this imper-
misclassification would exaggerate the long-term fect variable will not fully account for the con-
benefits of breastfeeding. founding effect of true income.

On the other hand, in nondifferential As discussed above, the direction of residual
misclassification, measurement error is inde- confounding may vary between high-income and
pendent of exposure or outcome status. This less developed settings. These differences will
leads to a dilution of the actual effect, because be explored when analysing the results of this
some breastfed subjects are classified as non- review for each outcome.
breastfed and vice-versa. Consequently, one is
less likely to detect an association, even if one Self-selection
really exists.
Even within the same social group, mothers who
Confounding by socioeconomic breastfeed are likely to be more health-conscious
status than those who do not breastfeed. This may also
lead them to promote other healthy habits
Socioeconomic status is one of the most im- among their children, including prevention of
portant confounders in studies on the long-term overweight, promotion of physical exercise and
effects of breastfeeding. In most societies, intellectual stimulation. This may be particu-
breastfeeding rates differ among social groups larly true in high-income populations. Because
(20). The direction of confounding by socio- these maternal attributes are difficult to meas-
economic status may vary between high-income ure, it is not possible to include them in the
and low/middle-income populations. In high-in- analyses as confounding factors. Nevertheless,
come countries, breastfeeding mothers tend to this possibility should be taken into account
be of higher educational and socioeconomic sta- when interpreting the study’s results.
tus (21); other things being equal, their offspring
will have a lower prevalence of cardiovascular Adjustment for potential mediating
risk factors and higher educational attainment factors
because they belong to the upper social classes.
Consequently, confounding by socioeconomic Several studies on the long-term consequences
status may overestimate the beneficial effects of breastfeeding have adjusted their estimates
of breastfeeding. for variables that may represent mediating fac-
tors, or links, in the causal chain leading from
On the other hand, in low/middle-income set- breastmilk to the outcomes. Adjustment for
tings breastfeeding is often more common

GENERAL METHODOLOGICAL ISSUES IN STUDIES OF THE LONG-TERM EFFECTS OF BREASTFEEDING 5

mediating factors will tend to underestimate the riod of the study cohorts’ births may affect the
overall effect of breastfeeding, e.g. adjusting for long-term effects of breastfeeding, being a source
weight at the time of blood pressure measure- of heterogeneity among the studies.
ment when evaluating the association between
breastfeeding and blood pressure in later life. Length of recall of breastfeeding
The “adjusted” estimate will reflect the residual
effect of breastfeeding which is not mediated Misclassification of breastfeeding duration has
by current weight (23). been discussed above. Feeding histories were
often assessed retrospectively and the length of
Main sources of heterogeneity recall has varied widely among studies. Accord-
among studies ing to the above-mentioned study by Huttly et
al (19) in southern Brazil, as many as 24% of
Current epidemiological practice places limited the mothers misclassified the duration of
value on the findings of a single study. Evidence is breastfeeding, and misclassification increased
built by pooling the results from several studies, with the time elapsed since weaning. Other stud-
if possible from different populations, either ies have also reported poor maternal recall of
through systematic reviews or meta-analyses. breastfeeding duration (27-29). As in Brazil,
Promislow et al (27) reported from the United
A major concern regarding systematic re- States that mothers who breastfed for a short
views and meta-analyses is the extent to which period were more likely to exaggerate
the results of different studies can be pooled. breastfeeding duration, while the opposite was
Heterogeneity of studies is unavoidable, and may observed for women who breastfed for long pe-
even be positive as it enhances generalizability. riods. Length of recall is therefore a potential
The question is not whether heterogeneity is source of heterogeneity among studies.
present, but if it seriously undermines the con-
clusions being drawn. Rigorous meta-analyses Source of information on
should incorporate a detailed investigation of breastfeeding duration
potential sources of heterogeneity (24). In the
present meta-analyses, the following possible The vast majority of the studies reviewed as-
sources of heterogeneity were considered for all sessed infant feeding by maternal recall, while
reviews. others relied on information collected by health
workers or on the subjects’ own reports. Mar-
Year of birth mot et al (30), in England, observed that about
65% of subjects correctly recalled whether they
Studies on the long-term effect of breastfeeding had been breastfed or formula-fed, and bottle-
have included subjects born during several dec- fed subjects were more likely to report wrongly
ades in the last century. During this period, the that they had been breastfed. If misclassification
diets of non-breastfed infants in now high-in- were independent of other factors related to
come countries have changed markedly. In the morbidity in adulthood, such as socioeconomic
first decades of the 20th centur y, most non- status, this misclassification would be non-dif-
breastfed infants received formulations based on ferential and would tend to underestimate the
whole cow’s milk or top milk (25), with a high long-term effects of breastfeeding.
sodium concentration and levels of cholesterol
and fatty acids that are similar to those in Categories of breastfeeding duration
mature breastmilk. By the 1950s, commercially
prepared formulas became increasingly popular. Among the reviewed studies, most compared
At this time, formulas tended to have a high ever-breastfed subjects to those who were never
sodium concentration and low levels of iron and breastfed. Other studies compared subjects
essential fatty acids. Only after 1980, the so- breastfed for less than a given number of months,
dium content was reduced and nowadays the often 2-3 months (including those who were
majority of formulas have levels that are similar never breastfed), to those breastfed for longer
to those in breastmilk (26). Therefore, the pe- periods. Few studies treated breastfeeding du-
ration as a continuous or ordinal variable with

6 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

several categories, thus allowing dose-response Among studies carried out in the last few
analyses. Furthermore, breastfeeding patterns decades, the type of milk fed to non-breastfed
(exclusive, predominant, or partial) have rarely infants would have varied substantially between
been assessed. high-income countries (where most babies re-
ceive industrialized formulas) and those from
Studies comparing ever versus never breast- low- and middle-income countries (where whole
fed subjects may be subject to misclassification. or diluted animal milk is often used).
The study by Huttly et al (19) showed that
mothers who had actually breastfed for up to 4 In this sense, lack of breastfeeding is an unu-
weeks often reported, at a later time, that they sual variable in epidemiological studies. For ex-
never breastfed. posures to, for example, smoking, alcohol or en-
vironmental risks, the reference category is
The comparison of ever versus never made up of those who are unexposed. In the
breastfed makes sense if the early weeks of life case of breastfeeding, however, those “unex-
are regarded as a critical period for the pro- posed” to it are themselves exposed to a number
gramming effect of breastfeeding on adult dis- of other foodstuffs, including animal milk, in-
eases (31). On the other hand, if there is no dustrialized or home-made formulas, or tradi-
critical window and breastfeeding has a cumu- tional weaning foods. Because alternative foods
lative effect, comparisons of ever versus never vary markedly from one setting to another, the
breastfed infants will lead to substantial under- effects of breastfeeding may be particularly af-
estimation of the effect of breastfeeding. fected by where the study was carried out. The
location (area, country) of the study is there-
Study setting fore a potential modifier of the effect of
breastfeeding.
Nearly all studies on the long-term consequences
of breastfeeding have been conducted in high- Adjustment for potential mediating
income countries and in predominantly Cauca- factors
sian populations. The findings from these stud-
ies may not hold for other populations exposed This issue has been discussed above. Inappro-
to different environmental and nutritional con- priate adjustment was investigated as a poten-
ditions, such as ethnic minorities in high-in- tial source of heterogeneity among studies.
come countries (32) or populations from less
developed countries.

GENERAL METHODOLOGICAL ISSUES IN STUDIES OF THE LONG-TERM EFFECTS OF BREASTFEEDING 7

III. Search methods

Selection criteria for studies Search strategy

In the present meta-analyses, we selected obser- In order to prevent selection bias (33) by cap-
vational and randomized studies, published in turing as many relevant studies as possible, two
English, French, Spanish or Portuguese, examin- independent literature searches were conducted:
ing the long-term effects of breastfeeding on the one at the Department of Child and Adolescent
following outcomes: blood pressure, overweight Health and Development in the World Health
or obesity, cholesterol, type-2 diabetes, and in- Organization (R.B.) and another at the Federal
tellectual performance. Studies restricted to University of Pelotas in Brazil (B.L.H.).
outcome measurement in infants were excluded
from the meta-analyses. Medline (1966 to March 2006) was searched
using the following terms for breastfeeding du-
Only those studies with internal compari- ration: breastfeeding; breast feeding; breastfed;
son groups were included. The type of compari- breastfeed; bottle feeding; bottle fed; bottle feed;
son group used (never breastfed, breastfed for infant feeding; human milk; formula milk; for-
less than x months, etc.) did not constitute an mula feed; formula fed; weaning.
eligibility criterion, but, as discussed above, the
way in which breastfeeding was categorized was We combined the breastfeeding terms, with
investigated as a potential source of heteroge- the following terms for each of the studied out-
neity among the studies. comes:

Types of outcome measures z Cholesterol: cholesterol; LDL; HDL;
triglycerides; or blood lipids.
According to the objectives of the present re-
view, we looked for studies with the following z Type-2 diabetes: diabetes; glucose; or glyc-
outcomes: emia.

z blood pressure: mean difference (in mmHg) z Intellectual performance: schooling; develop-
in systolic and diastolic blood pressure; ment; or intelligence.

z cholesterol: mean difference (in mg/dl) in z Blood pressure: blood pressure; hyperten-
total cholesterol; sion; systolic blood pressure; or diastolic
blood pressure.
z overweight and obesity: odds ratio compar-
ing breastfed and non-breastfed subjects; z Overweight or obesity: overweight; obesity;
body mass index; growth; weight; height;
z type-2 diabetes: odds ratio comparing child growth.
breastfed and non-breastfed subjects (or
alternatively, mean difference in blood In addition to the electronic search, the ref-
glucose levels); erence lists of the articles initially identified were
searched, and we also perused the Scientific
z intellectual performance: mean attained Citation Index for papers citing the articles iden-
schooling and performance scores in de- tified. Attempts were made to contact the au-
velopmental tests. thors of all studies that did not provide suffi-
cient data to estimate the pooled mean effects.
We also contacted the authors to clarify any
queries on the study’s methodology.

8 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

IV. Review methods

Assessment of study quality Data analysis

Eligible studies were evaluated for methodologi- Pooled effect estimate
cal quality prior to consideration of their re-
sults. The following a priori criteria for quality Effect measures were reported as i) weighted
assessment were used: mean differences and their 95% confidence in-
terval (CI) for continuous outcomes, and ii)
a. Losses to follow-up (%) pooled odds ratios and 95% CI for dichotomous
b. Type of study outcomes. Subjects were classified as either
breastfed or non-breastfed, according to the
(0) Observational specific classification used in each study. For
(1) Randomized the continuous outcomes, a negative mean dif-
c. Birth cohort ference denoted a lower value among breastfed
(0) No subjects, whereas for dichotomous outcomes an
(1) Yes odds ratio <1 denoted that breastfed subjects
d. Length of recall of breastfeeding duration presented lower odds of the outcome.
(0) >3 years
(1) <3 years Fixed or random-effects model
e. Source of information on breastfeeding
(0) Records Fixed-effect models assume that each study es-
(1) Interviews with subjects timates the same true population value for the
(2) Mothers effect of interest, and thus that differences be-
f. Control for confounding tween their results can be fully accounted for
(0) None by sampling variation; in this model, individual
(1) Socioeconomic or demographic studies are simply weighted by their precision
(34). On the other hand, random-effects mod-
variables els assume that population effects also vary, and
(2) Socioeconomic and demographic thus need to be accounted for as an additional
source of variation. The random-effects model
variables (35) gives greater weight to smaller studies, re-
(3) Socioeconomic, demographic sulting in a wider confidence interval than fixed-
effects models. In the latter, an important sta-
variables and maternal tistical question is whether variability among
anthropometr y studies is greater than would be expected with
g. Control for possible mediating variables the play of chance. In the present meta-analy-
(0) Yes ses, the Q-test was used to assess the heteroge-
(1) No neity among studies (36); if significant, the be-
tween-studies variability was higher than ex-
Data abstraction pected by chance, and this required the use of a
random-effects model (35).
Data from each study were extracted using a
standardized protocol to assess the mode of feed- Publication bias
ing, outcome, potential sources of heterogene-
ity and assessment of study quality. With re- Studies showing statistically significant associa-
spect to the assessment of study quality, each tions are more likely to appear in print, to be
study was independently evaluated by two re- published in English, and to be cited by others
viewers for each of the quality items, with disa-
greements resolved by consensus rating.

9

papers. Therefore, such articles are more likely to a fixed-effect. This analysis was performed
to be identified and included in a meta-analysis. using the METAREG command within STATA.
Publication bias is more likely to affect small In these random-effects regression (meta-regres-
studies, since the greater amount of time and sion) models, each of the items used to assess
money spent in larger studies makes them more study quality was considered as a covariate, in-
likely to be published, even when the results are stead of using an overall quality score. This al-
not statistically significant (34). lows the identification of aspects of the study
design, if any, which may be responsible for the
In the present meta-analyses, funnel plot and heterogeneity between studies (41). Further-
Egger’s test were employed to assess whether more, the following study characteristics were
there was any evidence of publication bias (37). also included as covariates in random-effects
We did not use the Begg test (38) because the regression:
regression method (Egger’s test) showed a bet-
ter performance for detection of funnel plot a. Definition of breastfeeding
asymmetry (39). Furthermore, we stratified the b Birth year
analyses according to study size, in order to as- c Age at outcome assessment:
sess the potential impact of publication bias on
the pooled estimate. (0) 1–9 years
(1) 10–19 years
Assessing heterogeneity (2) >19 years
d. Study size (n)
The next step after obtaining pooled results is e. Provenance (high-income country /
to assess whether certain study characteristics middle/low-income country).
may explain the variability between results. In
the present meta-analyses we used a random- As discussed in the Introduction, the present
effects regression model developed by Berkey et review was aimed at assessing the long-term
al (40) for evaluation of sources of heterogene- consequences of breastfeeding on five different
ity. In this approach, if the data are homogene- outcomes. This resulted in five separate meta-
ous or if heterogeneity is fully explained by the analyses which are described below.
covariates, the random-effects model is reduced

10 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

V. Results and discussion

Review 1 - Breastfeeding and blood pressure in later life

High blood pressure in adulthood is associated dren assigned to a low sodium diet in the first 6
with increased risk of ischaemic heart disease months of life. Therefore, there is no consensus
and stroke (42,43). It has been suggested that on whether the sodium content of infant diets
adult blood pressure is influenced by early life may lead to higher blood pressure in the future.
exposures, such as intrauterine growth, catch-
up growth, and infant feeding (44). Fatty acid content of breastmilk

Biological plausibility Long-chain polyunsaturated fatty acids are
present in breastmilk, but not in most brands
Three possible biological mechanisms for a pos- of formula (49); these substances are important
sible programming effect of breastfeeding on structural components of tissue membrane sys-
blood pressure have been proposed. tems, including the vascular endothelium (50).
Evidence suggests that dietary supplementation
Differences in sodium content be- with long-chain polyunsaturated fatty acids low-
tween breastmilk and formula ers the blood pressure in hypertensive subjects
(51). Furthermore, Forsyth et al (52) reported
As previously discussed (section on general that blood pressure at 6 years was lower among
methodological issues), until the 1980s the so- formula-fed children who had been assigned to
dium content of breastmilk in Western coun- a formula supplemented with long-chain poly-
tries was much lower than that of formulas (26). unsaturated fatty acids than among those
Because low sodium intake is related to lower randomized to a standard formula. This is, there-
blood pressure (45), it has been suggested that fore, a potential mechanism for a possible ef-
differences in sodium content between breast- fect of breastfeeding.
milk and formula would be one of the mecha-
nisms for the programming of later blood pres- Obesity
sure.
The effect of early infant feeding on blood pres-
However, evidence on the existence of an sure might also be mediated by overweight or
effect of early salt intake on later blood pres- obesity in adulthood, as this is a risk factor for
sure is controversial. Whitten & Stewart (46) hypertension (53). On the other hand, as dis-
reported that blood pressure at eight years of cussed below, the evidence suggests that
age was not correlated with sodium intake at breastfeeding has only a small protective effect
the age of eight months. In another study, against excess weight. Whether or not this small
Singhal et al (47) followed pre-term infants who effect may influence blood pressure levels re-
had been randomly assigned to receive two dif- mains to be proven.
ferent types of infant formulas which differed
greatly in salt content. Blood pressure at age In conclusion, of the three postulated mecha-
13-16 years was independent of the type of for- nisms, only the fatty acid content of breastmilk
mula the subject had received, but was lower appears to be supported by the literature, but
among the breastfed subjects. On the other there may well be other mechanisms that are
hand, Geleijnse et al (48) reported that adjusted currently unknown.
systolic blood pressure at age of 15 years was
3.6 mmHg (95% CI: -6.6 to -0.5) lower in chil-

11

Specific methodological issues The meta-analysis by Martin et al in 2005

General methodological issues affecting studies (55) included 15 studies that related breastfeed-
of the long-term consequences of breastfeeding ing to blood pressure measured after the age of
were addressed above. An additional issue is that 12 months. Meta-regression analysis was used
several studies included adjustment for current to evaluate whether the mean effect of
weight, body mass index or ponderal index in breastfeeding varied according to the subject’s
their multivariate analyses. Had breastfeeding age when the blood pressure was measured, year
been associated with adult weight, adjustment of birth, study size, length of recall of
for the latter would lead to an underestimation breastfeeding duration, follow-up rates, and
of its true effect on blood pressure. However, as whether or not confounding variables were ad-
discussed in section 1.3 above, there is no strong justed for.

evidence of a breastfeeding effect on adult Breastfeeding and systolic blood
weight, and therefore one would not expect ad- pressure
justment for weight to change the association
with blood pressure. Figure 1.1 shows that, in spite of the difference
in the admission criteria, the effect size in both

Overview of the existing meta- meta-analyses was similar, systolic blood pres-
analyses sure being significantly lower among breastfed
infants.
We reviewed two existing meta-analyses on the
influence of breastfeeding on blood pressure in However, there was evidence of publication
bias; the effect size decreased with increasing

later life (54,55). study size. Owen et al (54) noted that studies

The meta-analysis by Owen et al in 2003 (54) with fewer than 300 participants reported a

obtained information from 25 studies, includ- mean difference in systolic blood pressure of

-2.05 mmHg (95% CI: -3.30 to

-0.80) when comparing breastfed

Figure 1.1. Mean difference in systolic blood pressure between and non breastfed infants,
breastfed and non breastfed subjects whereas among studies with

Mean difference (95% CI) in systolic more than 1000 subjects the
blood pressure (mm Hg)
0 mean difference was -0.16

mmHg (95% CI: -0.60 to 0.28).
-0.5

Martin et al (55) also reported

-1 similar differences in mean ef-

fect according to study size.
-1.5

In terms of heterogeneity

-2 among studies, the magnitude of

-2.5 Martin et al (55) the effect was independent of the
Ow en et al (54) subjects’ age. On the other hand,
Martin et al (55) observed that
Meta-analysis the protective effect of

breastfeeding was higher among

ing those in which blood pressure was meas- those born in or before 1980
ured in infancy. Meta-regression analysis was (mean difference: -2.7 mmHg) compared with
used to evaluate whether there were differences those born after (mean difference: -0.8 mmHg),
in the mean effect of breastfeeding according whereas Owen et al (54) failed to observe such
to the subject’s age, year of birth, study size, an association.
length of recall for information on breastfeeding
duration (for retrospective studies), and adjust- Martin et al (55) noticed that only 6 of the
ment for current body size. 15 studies included in their meta-analysis had
controlled for confounding by socioeconomic
status and maternal variables (body mass index,

12 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

smoking). In a clear demonstration of the im- Unlike for systolic blood pressure, there was
portance of adjustment for confounding, in two no evidence of publication bias. The mean ef-
of the three studies that reported crude and fect of breastfeeding on diastolic blood pres-
adjusted estimates, adjustment reduced the sure was similar among studies with 1000 or
crude estimates in more than 30% (Table 1.1). more participants (mean difference: -0.4 mmHg;
With regard to control for possible mediating 95% CI: -0.9 to 0.1) and smaller studies (<1000
factors, Owen et al (54) reported that control participants) (mean difference: -0.6 mmHg; 95%
for body size at the time of blood pressure as- CI: -1.5 to 0.2).

Table 1.1. Studies included in the Martin et al (55) meta-analysis that provided crude and adjusted esti-
mates of difference in systolic blood pressure between breastfed and non breastfed subjects

Study Mean difference in systolic blood Covariates included in the multivariate model
Martin 2004 (56) pressure in mm Hg (SE)
Sex, age, room temperature, solids introduced, maternal
Crude Adjusted factors (schooling, social class, age at birth of the child,
hypertension, pre-pregnancy BMI, height), paternal BMI,
-1.2 (SE* 0.4) -0.8 (SE 0.4) child's height

Martin 2005 (57, 58) -0.33 (SE 1.1) -0.11 (SE 1.1) Age, birth order, father's social class, social class in
Lawlor 2004 (59) -1.0 (SE 0.3) -1.2 (SE 0.4) adulthood

Age, sex, parental factors (age, BMI, smoking in
pregnancy, schooling, marital status), family income,
birthweight, adiposity

* SE: Standard error of mean difference

sessment had no effect in the results of 10 stud- Age and year of birth were not related to

ies (12 observations) that included such analy- variability in the study results. On the other

ses. This is in agreement with our earlier argu- hand, those studies that relied on maternal re-

ment that – given the weak association between call of breastfeeding beyond infancy showed

breastfeeding and adult weight – such control smaller differences between breastfed and for-

would not affect the final results. mula-fed groups than studies with shorter re-

Breastfeeding and diastolic call.
blood pressure
Figure 1.2. Mean difference in diastolic blood pressure between
Figure 1.2 shows that in the meta- breastfed and non breastfed subjects

analysis by Owen et al (54) the dif-

ference in diastolic blood pressure Mean difference (95% CI) in diastolic 0.2
blood pressure (mm Hg)
between breastfed and non- 0
breastfed infants was not statisti- -0.2
cally significant (mean difference:

-0.36 mmHg; 95% CI: -0.79 to 0.08 -0.4
mmHg). On the other hand, Mar- -0.6
tin et al (55) found that mean

diastolic blood pressure was signifi- -0.8

cantly lower among breastfed infants -1 Martin et al (55)
(mean difference: -0.5 mmHg; 95% Ow en et al (54)
CI: -0.9 to -0.04 mmHg).
Meta-analysis

RESULTS AND DISCUSSION - REVIEW 1 13

Studies not included in the pre- lished meta-analyses, and those identified by the
vious meta-analyses two independent literature searches at WHO
and at the University of Pelotas. It was possible
We have identified four recently published stud- to include 30 estimates on the effect of
ies that have not been included in the published breastfeeding on systolic blood pressure, and 25
meta-analyses. Below are summarized the find- on diastolic blood pressure (Table 1.2). Fig. 1.3
ings from these studies. and 1.4 show the forest plot for systolic and
diastolic blood pressures, respectively. Systolic
Martin et al (57) studied a cohort of 1580 (mean difference: -1.21 mmHg; 95% CI: -1.72
men living in Caerphilly, South Wales, who were to -0.70) and diastolic blood pressures (mean
aged 45-59 years when examined between 1979 difference: -0.49 mmHg; 95% CI: -0.87 to -0.11)
and 1983. Information on breastfeeding dura- were both lower among those subjects who had
tion was obtained from the subjects’ mothers been breastfed. Random-effects models were
or a close female relative. Difference in systolic used in both analyses because heterogeneity
blood pressure between breastfed and bottle-fed among studies was statistically significant.
subjects was -0.11 mmHg (95% CI: -2.28 to 2.06)
and for diastolic blood pressure -0.21 mmHg Similar to the previously published meta-
(95% CI: -1.67 to 1.25). analyses, publication bias was clearly present.
Table 1.2 shows that the mean difference was
Martin et al also studied (58) a historic co- inversely related to the study size, with larger
hort based on a follow-up of the subjects who studies reporting a smaller protective effect of
participated in a 1-week survey of diet and health breastfeed-ing. This was more marked for
when aged 0 to 19 years, between 1937 and systolic than for diastolic blood pressure. This
1939. This study was conducted in 16 centres is confirmed by examination of the funnel plots
in England and Scotland. Information on which are clearly asymmetrical, with small stud-
breastfeeding duration was obtained from the ies reporting a higher protective effect of
subject’s mother at the time of the survey on breastfeeding. (Fig. 1.5 and 1.6)
diet and health. Differences between breastfed
and non-breastfed subjects were -1.62 mmHg Conclusion
(95% CI: -6.66 to 3.41) and -0.74 mmHg (95%
CI: -3.06 to 1.57) for systolic and diastolic blood According to Owen et al (54), the association
pressure, respectively. between breastfeeding and lower blood pressure
was mainly due to publication bias, and any ef-
Lawlor et al (60) evaluated random samples fect of breastfeeding was modest and of limited
of schoolchildren aged 9 years and similar sam- clinical or public health relevance. In spite of
ples of 15-year-olds from Estonia (n=1174) and not being able to exclude residual confounding
Denmark (n=1018). Even after controlling for and publication bias, Martin et al (55) concluded
possible confounding variables, the systolic blood that breastfeeding was negatively associated with
pressure was lower among those children who blood pressure. They argued that even a small
had ever been exclusively breastfed (difference: protective effect of breastfeeding would be im-
-1.7 mmHg; 95% CI: -3.0 to -0.5). portant from a public health perspective. For
example, a reduction in mean population blood
In a recently published paper, Horta et al pressure of 2 mmHg could lower the prevalence
(61) observed in a cohort of over 1000 15-year- of hypertension by up to 17%, the number of
olds in Pelotas (Brazil) that breastfeeding was coronary heart disease events by 6%, and stroke
not related to systolic (difference -1.31 mmHg; by 15%. Three large studies were published since
95% CI: -3.92 to 1.30) or diastolic blood pres- the last review, two of which found no associa-
sure (difference -0.64 mmHg; 95% CI: -2.91 to tion and one found a protective effect of
1.63). breastfeeding.

Update of existing meta-analysis Both meta-analyses may have been affected
by publication bias. Small studies with negative
A new meta-analysis was carried out which in- results are less likely to be published, and this
cluded the four recently published studies de-
scribed above, all the papers in previously pub-

14 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Table 1.2. Breastfeeding and blood pressure in later life: studies included in the meta-analysis in ascending order of subjects’ age at which outcome was measured.

Author, year Study design Year of birth Age at blood Gender Comparison groups Mean difference in mm Hg (Standard
(reference) of subjects pressure error of mean difference)
measurement

Systolic blood Diastolic blood
pressure pressure

Boulton 1981(62) Cross-sectional 1976-9 1 year All Breastfed for ≥3 months (n=14) vs. Formula fed -4.11 (SE 3.12) Not stated
from <3 months (n=47)

Smith 1995 (63) Cohort 1990 1 year All Ever breastfed vs. Never breastfed -2.29 (SE 1.49) Not stated
Not stated 8-26 months
Zeman 1981 (64) Cross-sectional 1981-3 3-4 years All Breastfed (n=42), vs. Bottle fed (n=13) -4.91 (SE 2.10) 0.33 (SE 0.90)

Baronowski 1992 (65) Cross-sectional All Exclusively breastfed for >3 months (n=60) vs. -2.17 (SE 1.61) 0.48 (SE 1.42)
Bottle fed (n=185)

Lawlor 2004 (59) Cohort 1981-4 5 years All Breastfed for ≥6 months (n=1191) vs. Breastfed -1.19 (SE 0.4) Not stated

for <6 months (n=2528)

Whincup 1989 (66) Cross-sectional 1979-83 5-8 years All Exclusively breastfed for first 3 months -0.20 (SE 0.33) -0.26 (SE 0.23)
(n=1221) vs. Bottle fed (n=1787)

Forsyth 2003 (52) Cohort 1992 5 years All Breastfed (n=83) vs. Formula fed (n=71) -2.2 (SE 1.6) -3.4 (SE 1.4)
Williams 1992 (67) Cohort 1972-3 7 years
All Exclusively breastfed {median 28 wk} (n=192) -0.5 (SE 0.66) -0.7 (SE 0.61)

vs. Bottle fed (n=327)

Wilson 1998 (68) Cohort 1983-6 6-9 years All Exclusively breastfed for >15 wk (n=74) vs. -2.84 (SE 1.40) -1.79 (SE 1.23)

Bottle fed (n=105)

Martin 2004 (56) Cohort 1991-2 6-9 years All Ever breastfed (n=5586) vs. Never breastfed -0.8 (0.4) -0.6 (0.3)

(n=1141)

RESULTS AND DISCUSSION - REVIEW 1 Lucas 1994 (69) Randomized controlled 1982-5 7-8 years All Allocated to banked breastmilk (n=66) vs. 0.1 (SE 1.55) -0.7 (SE 1.32)
trial 8-9 years Allocated to preterm formula (n=64)
Rona 1996 (70) 8-9 years
Cross-sectional 1983-5 8-9 years All Exclusively breastfed for ≥3 mo (n=157) vs. 1.46 (SE 0.84) 2.54 (SE 0.76)
Rona 1996 (70) 9-11 years
(England 1993 sample) Bottle fed (n=308)
Rona 1996 (70)
Cross-sectional 1984-6 All Exclusively breastfed for ≥3 mo (n=213) vs. -2.75 (SE 0.86) -2.34 (SE 0.75)
Esposito-Del Puente
1994 (71) (England 1994 sample) Bottle fed (n=318)

Cross-sectional 1983-6 All Exclusively breastfed for ≥3 mo (n=124) vs. -1.29 (SE 1.01) -0.97 (SE 0.84)

(Scotland 1994 sample) Bottle fed (n=273)

Cross-sectional 1980-2 All Breastfed (n=43) vs. Bottle fed (n=17) -4.13 (SE 2.86) 0.11 (SE 0.86)

15

16 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES Table 1.2. (continued) Year of birth Age at blood Gender Comparison groups Mean difference in mm Hg (Standard
Author, year (reference) Study design of subjects pressure error of mean difference)
measurement

Systolic blood Diastolic blood
pressure pressure

Lawlor 2005 (60) Cross-sectional Not stated 9-15 years All Exclusively breastfed vs. Not exclusively -1.7 (SE 0.66) Not stated
All breastfed
British Cohort Study (72) Cohort 1970 10 years All -0.16 (SE 0.38) -0.21 (SE 0.33)
All Breastfed for ≥3 months (n=951) vs. Never
Singhal 2001 (47) Randomized 1982-5 13-16 years All breastfed (n=5133) -2.7 (SE 1.52) -3.2 (SE 1.31)
Owen 2002 (54) controlled trial All
1982-6 13-16 years All Allocated to banked breastmilk (n=66) vs. 0.32 (SE 0.64) -0.17 (SE 0.35)
Cross-sectional All Allocated to preterm formula (n=64)
All
Taittonen 1996 (73) Cohort 1962-74 12-24 years All Exclusively breastfed ≥3 months (n=980) vs. -5.48 (SE 1.14) Not stated
All Exclusively bottle fed (n=951)
Horta 2006 (61) Cohort 1982 15 years All -1.31 (SE 1.33) -0.64 (SE 1.16)
Williams 1992 (67) Cohort 1972-3 18 years Female Breastfed >3 months (n=760) vs. Bottle fed -2.63 (SE 1.34) -2.66 (SE 1.36)
Male (n=150)
Kolacek 1993 (74) Cohort 1968-9 18-23 years All 0.00 (SE 1.96) 0.75 (SE 1.40)
Breastfed (n=966) vs. Bottle fed (n=85)
Leeson 2001 (75) Cross-sectional 1969-75 20-28 years 0.0 (SE 1.55) -1.0 (SE 0.94)
Martin 2003 (76) Cohort 1972-4 23-27 years Exclusively breastfed {median 28 wk} (n=197) -3.7 (SE 1.38) -1.4 (SE 0.97)
or bottle fed (n=319)
Ravelli 2000 (77) Cohort 1943-7 48-53 years 0.2 (SE 1.65) 0.9 (SE 1.08)
Exclusively breastfed ≥3 months (n=199) vs.
Wadsworth 1987 (78) Cohort 1946 53 years Never breastfed (n=78) -0.94 (SE 0.97) -0.32 (SE 0.37)

Fall 1995 (79) Cohort 1920-30 60-71 years Breastfed (n=149) vs. Bottle fed (n=182) -2.64 (SE 2.88) -0.96 (SE 1.38)

Martin 2005 (57, 58) Cohort 1920-38 45-59 years Lowest (n=193) vs. Highest quartile (n=79) of -0.11 (SE 1.11) -0.21 (SE 0.74)
Martin 2005 (58) Cohort 1918-39 63-82 years dried formula milk consumption at age 3 months -1.62 (SE 2.57) -0.74 (SE 1.18)

Exclusively breastfed (n=520) vs. Bottle fed
(n=105)

Exclusively breastfed for ≥3 months (n=1507)
vs. Never breastfed (n=628)

Exclusively breastfed (n=862) vs. Bottle fed
(n=68)

Breastfed (n=1159) vs. Bottle fed (n=417)

Breastfed (n=272) vs. Bottle fed (n=90)

Figure 1.3. Mean difference in systolic blood pressure in mm Hg (and its 95% confidence interval) between
breastfed and non-breastfed subjects in different studies. Whether the estimate was for males
(M), females (F) and all (A) is indicated in parenthesis

Boulton 1981 (M) 11 1122 12 11112222
Zeman 1981 (A) 11 1 1122
Baronowski 1992 (A) 12 11
Whincup 1989 (A) 1 11112222 11
Forsyth 2003 (A) 1122
Williams 1992 (A) 11
Wilson 1998 (A) 1122
Martin 2004 (A)
Lucas 1994 (A) 11 111222
1122
Rona 1996 (A)
Rona 1996 (A) 1122
Rona 1996 (A) 1122
Esposito-Del Puente 1994 (A) 1
British Cohort Study (A)
Singhal 2001 (A) 11 11
Owen 2002 (A) 1122 11
Taittonen 1996 (A)
Williams 1992 (A) 11
Kolacek 1993 (A) 1122 111222
Leeson 2001 (A)
Martin 2003 (A) 1122
Ravelli 2000 (A) 12
Wadsworth 1987 (A)
11223344556677
Fall 1995 (F)
Martin 2005 (A)
Martin 2005 (M)
Lawlor 2005 (A)
Lawlor 2004 (A)
Horta 2006 (A)
Smith 1995 (A)

Combined

-10 -5 -1 0 15

Mean difference (mm Hg) Mean lower in breastfed Mean higher in breastfed
subjects subjects

leads to an overestimate of the pooled mean founding may vary according to the level of eco-
difference due to selective inclusion of small nomic development of the population.
positive studies.
In summary, the present updated meta-analy-
Lack of control for confounding is another ses show that there are small but significant pro-
methodological issue, as pointed out by Martin tective effects of breastfeeding on systolic and
et al (55). Most studies did not provide esti- diastolic blood pressure. Publication bias is un-
mates adjusted for confounding by socioeco- likely to explain this finding because a signifi-
nomic status and maternal characteristics; in cant protective effect was observed even among
the studies that showed both adjusted and crude the larger studies. However, residual confound-
results, the latter tended to overestimate the ing cannot be excluded because of the marked
protective effect of breastfeeding. The majority reduction in effect size after adjustment for
of these studies are from developed countries, known confounders.
and as previously discussed, the direction of con-

RESULTS AND DISCUSSION - REVIEW 1 17

Figure 1.4. Mean difference in systolic blood pressure in mm Hg (and its 95% confidence interval) between
breastfed and non-breastfed subjects in different studies. Whether the estimate was for males
(M), females (F) and all (A) is indicated in parenthesis

Zeman 1981 (A)
Baronowski 1992 (A)

Whincup 1989 (A)
Forsyth 2003 (A)
Williams 1992 (A)
Wilson 1998 (A)

Martin 2004 (A)
Lucas 1994 (A)
Rona 1996 (A)
Rona 1996 (A)
Rona 1996 (A)
Esposito-Del Puente 1994 (A)
British Cohort Study (A)
Singhal 2001 (A)
Owen 2002 (A)
Williams 1992 (A)
Kolacek 1993 (A)
Leeson 2001 (A)
Martin 2003 (A)
Ravelli 2000 (A)
Wadsworth 1987 (A)

Fall 1995 (F)
Martin 2005 (A)
Martin 2005 (M)
Horta 2006 (A)

Combined

-5 -1 0 1 5
Mean difference (mm Hg)
Mean lower in breastfed Mean higher in breastfed
subjects subjects

Figure 1.5. Funnel plot showing mean difference Figure 1.6. Funnel plot showing mean difference
in systolic blood pressure (mm Hg) by in diastolic blood pressure (mm Hg) by
standard error of mean difference standard error of mean difference

00

Standard error 1 0.5

2 1

3 -5 0 1.5 -2 0 2 4
-10 Mean difference (mm Hg) 5 -4 Mean difference (mm Hg)

18 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Table 1.3. Breastfeeding and blood pressure in later life: Random-effects meta-analyses by subgroups

Subgroup analysis Systolic blood pressure Diastolic blood pressure

Number of Mean difference (95% P value Number of Mean difference (95% P value
estimates confidence interval) estimates confidence interval)

By age group 14 -1.06 (-1.70 to -0.42) 0.04 11 -0.54 (-1.27 to 0.19) 0.14
1 to 9 years 6 -0.53 (-1.22 to 0.17) 0.14
9 to 19 years 8 -1.81 (-3.13 to -0.50) 0.007 8 -0.38 (-0.90 to 0.15) 0.16
>19 years
8 -0.95 (-1.97 to 0.07) 0.07

By study size

<300 participants 10 -2.45 (-3.52 to -1.38) 0.001 9 -0.87 (-1.83 to 0.08) 0.07

300-999 participants 11 -1.52 (-2.80 to -0.23) 0.02 9 -0.59 (-1.71 to 0.52) 0.30

≥1000 participants 9 -0.59 (-1.00 to -0.19) 0.004 7 -0.33 (-0.61 to -0.04) 0.02

By year of birth of 13 -1.42 (-2.44 to -0.39) 0.007 11 -0.40 (-0.78 to -0.02) 0.04
subjects
16 -1.03 (-1.61 to -0.45) 0.001 13 -0.61 (-1.27 to 0.04) 0.06
Before 1980
After 1980

By length of recall of 19 -1.23 (-1.63 to -0.82) 0.001 14 -0.59 (-0.94 to -0.23) 0.001
breastfeeding
11 -1.01 (-1.97 to -0.06) 0.04 10 -0.36 (-1.02 to 0.30) 0.28
<3 years
≥3 years

By categorization of 14 -0.83 (-1.35 to -0.32) 0.002 13 -0.49 (-0.89 to -0.10) 0.01
breastfeeding
16 -1.42 (-2.22 to -0.63) 0.001 12 -0.47 (-1.10 to 0.16) 0.14
Ever breastfed
Breastfed for a
given number of
months

By control for 11 -1.73 (-3.17 to -0.30) 0.02 9 -0.59 (-1.39 to 0.22) 0.15
confounding 12 -0.92 (-1.71 to -0.14) 0.02 12 -0.44 (-1.04 to 0.17) 0.16

None 7 -1.19 (-1.70 to -0.69) 0.001 4 -0.61 (-1.12 to -0.10) 0.02
Adjusted for
socioeconomic
status
Adjusted for
socioeconomic and
demographic
variables

Study setting

High-income country 26 -1.15 (-1.68 to -0.62) 0.001 22 -0.54 (-0.95 to -0.14) 0.009

Middle/Low-income 4 -1.93 (-3.61 to -0.26) 0.02 3 0.13 (-1.12 to 1.37) 0.20

country

Total 30 -1.21 (-1.72 to -0.70) 25 -0.49 (-0.87 to -0.11)

RESULTS AND DISCUSSION - REVIEW 1 19

Review 2 - Breastfeeding and blood cholesterol in later life

Concentrations of total cholesterol and LDL derestimation of its true effect on cholesterol.
cholesterol (low-density lipoprotein) are impor- However, as discussed in the review on the long-
tant risk factors for coronary heart disease (80). term effects of breastfeeding on blood pressure,
It has been suggested that total cholesterol and the evidence suggests that the effect of
LDL cholesterol may be programmed by early breastfeeding on adult weight is weak, and there-
life exposures, such as rapid early growth (81) fore one would not expect an adjustment for weight
and infant feeding (82). to change the association with cholesterol.

Biological plausibility Overview of the evidence

The cholesterol content is markedly higher in A previous meta-analysis (82) on this associa-
breastmilk than in most commercially available tion showed that mean total cholesterol in in-
formulas. High cholesterol intake in infancy may fancy was higher among those breastfed (mean
have a long-term programming effect on syn- difference: 0.64; 95% CI: 0.50 to 0.79 mmol/L),
thesis of cholesterol by down-regulation of he- whereas among adults the total cholesterol was
patic hydroxymethylglutaryl coenzyme A (HMG- lower among those who had been breastfed
CoA) (83). This hypothesis is supported by stud- (mean difference -0.18; 95% CI: -0.30 to -0.06
ies with animals, which showed that early expo- mmol/L).
sure to increased levels of cholesterol is associ-
ated with decreased cholesterol levels at a later The electronic search carried out at WHO
age. Indeed, Devlin et al (84) reported that yielded 37 potentially relevant publications, 23
HMG-CoA reductase was higher (P < .05) in of which provided data on the mean difference
formula-fed than in milk-fed piglets, whereas LDL in total cholesterol between breastfed and non-
receptor mRNA was not independent of early breastfed subjects. From these 23 studies, 28
diet. HMG-CoA is the rate-limiting enzyme in estimates of total cholesterol were derived, of
synthesis of cholesterol from acetate, and HMG- which 18 included both genders and two were
CoA reductase inhibitors, the so-called statins, gender specific. No additional studies were iden-
have an important cholesterol-lowering effect tified by the independent search at the Univer-
(85). sity of Pelotas. Table 2.1 presents a description
of the studies included in the present meta-analy-
Therefore, nutritional programming by the sis, and Figure 2.1 shows the distribution of the
high cholesterol content of breastmilk has been studies’ results. There was strong evidence of
proposed as a potential mechanism for the as- heterogeneity between studies (P = 0.003), and
sociation between breastfeeding duration dur- the mean difference using a random-effects
ing infancy and lower cholesterol levels in later model was -0.03 (95% CI: -0.10 to 0.03), sug-
ages. gesting no overall association between
breastfeeding and cholesterol levels.
Specific methodological issues
However, Table 2.2 shows marked effect
General methodological issues affecting studies modification by age group. In adults (>19 years),
of the long-term consequences of breastfeeding breastfed subjects had mean total cholesterol
were addressed in the Introduction. An addi- levels 0.18 mmol/L (95% CI: 0.06 to 0.30 mmol/
tional issue affecting studies of cholesterol lev- L) lower than those who were bottle-fed, and
els is that four studies included in their there was no heterogeneity between studies (P
multivariate analyses an adjustment for weight, = 0.86). For children and adolescents, the asso-
body mass index or ponderal index measured at ciation was not significant. Figure 2.2 shows the
the same time as cholesterol levels. Had forest plot for studies in adults.
breastfeeding been associated with adult weight,
adjustment for the latter would lead to an un- Other subgroup analyses were carried out.
Studies with a length of recall of breastfeeding
duration >3 years resulted in lower mean total

20 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Table 2.1. Breastfeeding and blood cholesterol in later life: studies included in the meta-analysis in ascending order of subjects' age at which outcome was measured

Author, year Study design Year of birth Age at Gender Comparison groups Mean difference in
(reference) of subjects cholesterol total cholesterol
measurement mmol/L (SE)
-0.07 (SE 0.09)
Jooste 1991 (86) Cohort 1981-6 1 year All Exclusively breastfed (n=110) vs. Formula fed (n=201) for the first 3
months 0.35 (SE 0.21)
Mize 1995 (87) Cohort Not stated 1 year All 0.03 (SE 0.02)
Freedman 1992 (88) Cross-sectional 1972-90 1-4 years All Ever breastfed (n=23) vs. Formula fed from birth (n=33) 0.02 (SE 0.19)
Friedman 1975 (89) Cross-sectional Not stated 1.5-2 years All 0.25 (SE 0.13)
Ward 1980 (90) Cross-sectional 1974 2.5-3 years All Ever breastfed (n=2232) vs. Bottle fed (n=1796) 0.12 (SE 0.06)
Routi 1997 (91) Cohort 1990-2 3 years All
Breastfed (n=31) vs. Bottle fed (n=55) 0.2 (SE 0.11)
Huttunen 1983 (92) Cohort 1975 5 years All
Breastfed for ≥1 month (n=28) vs. Bottle fed (n=46) -0.57 (SE 0.21)
Elaraby 1985 (93) Cross-sectional Not stated 5-10 years All -0.3 (SE 0.13)
1981-7 5-11 years Male Breastmilk only source of milk at 7 months (n=295) vs. Formula only 0 (SE 0.16)
Plancoulaine 2000 (94) Cross-sectional Female source of milk at 7 months (n=324) -0.04 (SE 0.20)
All
Crawford 1981 (95) Cohort 1969-70 6 years Exclusively breastfed for 6 months (n=35) vs. Formula feeding 0.54 (SE 0.34)
Male started between 1 to 6 months (n=32) 0.23 (SE 0.27)
Hodgson 1976 (96) Cohort 1962-5 7-12 years Female
Breastfed for >6 months (n=50) vs. Artificially fed (n=50) -0.13 (SE 0.09)
RESULTS AND DISCUSSION - REVIEW 2 Fomon 1984 (97) Cohort 1966-71 8 years Male 0.03 (SE 0.12)
Female Breastfed (n=111) vs. Formula fed (n=131) -0.09 (SE 0.27)
Hromodova 1997 (98) Cohort Not stated 12-13 years All Breastfed (n=101) vs. Formula fed (n=118) -0.02 (SE 0.04)
1982-6 13-16 years All
Owen 2002 (82) Cross-sectional Exclusively breastfed (n=32) vs. Exclusively formula-fed (n=16) at 3 -0.4 (SE 0.20)
All months
21 Singhal 2004 (99) Randomized controlled 1982-5 13-16 years 0.03 (SE 0.25)
Friedman 1975 (89) trial Not stated 15-19 years All Exclusively breastfed (n= 12) vs. Formula fed (n=10) for the first 3
months
Cross-sectional Exclusively breastfed (n= 18) vs. Formula fed (n=10) for the first 3
months

Breastfed (n=70) vs. Formula fed (148)
Breastfed (n=76) vs. Formula fed (105)

Breastfed (n=44) vs. Formula fed (n=13)

Exclusively breastfed (n= 620) vs. Exclusively formula fed (n=542)
for the first 3 months

Allocated to banked breastmilk (n=66) vs. Allocated to preterm
formula (n=64)

Breastfed (n=86) vs. bottle fed (n=94)

Mean difference in
total cholesterol
mmol/L (SE)
-0.2 (SE 0.32)
-0.2 (SE 0.31)
-0.18 (SE 0.11)
-0.1 (SE 0.25)
-0.5 (SE 0.23)
-0.16 (SE 0.11)
0.12 (SE 0.28)
-0.32 (SE 0.25)
0.12 (SE 0.43)
cholesterol for subjects who were breastfed, al-
though the difference was not statistically sig-Gender Comparison groupsExclusively breastfed (n=35) vs. Formula fed (n=17)
nificant. Adjustment for body size at the timeExclusively breastfed (n=52) vs. Formula fed (n=14)
of cholesterol assessment was a source of het-Any breastfeeding (n=149) vs. Exclusively bottle fed (n=182)
erogeneity between studies; the protective ef-Exclusively breastfed for 5 months (n=57) vs. Exclusively formula
fect of breastfeeding was restricted to studiesfed (n=20)
that adjusted for body size (mean difference:Exclusively breastfed for 5 months (n=68) vs. Exclusively formula
-0.20; 95% CI: -0.33 to -0.06). Both types offed (n=27)
heterogeneity, however, could be explained byExclusively breastfed (n=520) vs. Partly or exclusively bottle fed
the age ranges of the study subjects because(n=105) in the first 10 days
longer recall and adjustment for current size wereBreastfed (n=272) vs. Bottle fed (n=90)
more frequent in studies of adults. Age at as-Exclusively breastfed (n=344) vs. Formula fed (n=25)
sessment of serum cholesterol explained 59.7%Exclusively breastfed (n=208) vs. Formula fed (n=11)
of the overall heterogeneity; further adjustment
for body size and length of recall did not pro-Male
vide further explanation for heterogeneity in theFemale
random-effects model.All
Male
Concerning publication bias, the funnel plotFemale
is quite symmetrical, with small studies tend-All
ing to report either positive or negative effectsMale
of breastfeeding, with no evidence of bias. In-Male
deed, Egger’s test was not statistically signifi-Female
cant (P=0.16). Furthermore, Table 2.2 shows
that the mean difference in total cholesterol wasAge at18-23 years20-28 years48-53 years
independent of study size.cholesterol31-32 years63-82 years
measurement 59-70 years
Conclusion 60-71 years

This meta-analysis suggests that the associationYear of birth1968-91969-751943-7
between breastfeeding and total cholesterol var-of subjects19461918-39
ies according to age. Whereas no significant1920-30
effect was observed in children or adolescents,1920-30
mean cholesterol levels among adults who were
Table 2.1. (continued) breastfed were 0.18 mmol/L (6.9 mg/dl) lowerStudy designCohortCross-sectionalCohort
than among non-breastfed subjects. This asso-CohortCohort
ciation did not seem to be due to either publi-Cohort
cation bias or confounding. The median level ofCohort
cholesterol in the adult studies included in the
review was about 5.7 mmol/L; the observed re-Author, yearKolacek 1993 (100)Leeson 2001 (75)Ravelli 2000 (77)
duction associated with breastfeeding corre-(reference)Marmot 1980 (30)Martin 2005 (58)
sponds to about 3.2% of this median.Fall 1992 (101)
Fall 1995 (79)
22 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Figure 2.1. Mean difference in total cholesterol in mmol/L (and its 95% confidence interval) between
breastfed and non-breastfed subjects in different studies. Whether the estimate was for males
(M), females (F) and all (A) is indicated in parenthesis

Jooste 1991 (A)
Mize 1995 (A)

Freedman 1992 (A)
Friedman 1975 (A)

Ward 1980 (A)
Routi 1997 (A)
Huttunen 1983 (A)
Plancoulaine 2000 (M)
Plancoulaine 2000 (F)
Crawford 1981 (A)
Hodgson 1976 (M)
Hodgson 1976 (F)
Hromodova 1997 (A)
Owen 2002 (A)
Friedman 1975 (A)
Kolacek 1993 (M)
Kolacek 1993 (F)
Leeson 2001 (A)
Marmot 1980 (M)
Marmot 1980 (F)
Ravelli 2000 (A)
Martin 2005 (A)

Fall 1992 (M)
Fall 1995 (F)
Singhal 2004 (A)
Elaraby 1985 (A)
Fomon 1984 (M)
Fomon 1984 (F)

Combined

-5 -1 0 1 5
Mean difference (mmol/L)
Mean lower in breastfed Mean higher in breastfed
subjects subjects

Figure 2.2. Mean difference in total cholesterol in mmol/L (and its 95% confidence interval) between
breastfed and non-breastfed subjects during adult life. Whether the estimate was for males
(M), females (F) and all (A) is indicated in parenthesis

Kolacek 1993 (M) 112233 111222333
Kolacek 1993 (F) 1111111122222222333333331144444225555533 1122
Leeson 2001 (A) 111222333 111112222233333444445555566666
Marmot 1980 (M) 111222333
Marmot 1980 (F)
1122334455667788
Ravelli 2000 (A)
Martin 2005 (A)

Fall 1992 (M)
Fall 1995 (F)

Combined

-1.5 -1 -0.5 0 0.5 1 1.5
Mean difference (mmol/L)
Mean lower in breastfed Mean higher in breastfed
subjects subjects

RESULTS AND DISCUSSION - REVIEW 2 23

Figure 2.3. Funnel plot showing mean difference in total choles-
terol (mmol/L) by standard error of mean difference

0

1

Standard error 2

3

4 -0.5 0 0.5 1
-1

Mean difference (mmol/L)

Table 2.2. Breastfeeding and blood cholesterol in later life: Random-effects meta-analyses of cho-
lesterol levels by subgroup

Subgroup analysis Number of estimates Mean difference (95% P value
of total cholesterol confidence interval)
By age group 0.63
1 to 9 years 15 0.02 (-0.06 to 0.11) 0.37
9 to 19 years 4 -0.07 (-0.21 to 0.08) 0.004
>19 years 9 -0.18 (-0.30 to -0.06)
0.47
By study size 20 -0.04 (-0.16 to 0.07) 0.74
<300 participants 8 -0.01 (-0.08 to 0.06)
≥300 participants 0.32
17 -0.06 (-0.18 to 0.06) 0.64
By year at birth 7 -0.02 (-0.10 to 0.06)
Before 1980 0.88
After 1980 18 -0.01 (-0.10 to 0.09) 0.35
9 -0.05 (-0.14 to 0.05)
By study design 0.95
Cross-sectional 21 0.00 (-0.07 to 0.08) 0.07
Cohort 7 -0.13 (-0.27 to 0.01)
0.08
By length of recall of breastfeeding 17 -0.07 (-0.16 to 0.01) 0.82
<3 years 11 0.01 (-0.11 to 0.13)
≥3 years 0.45
23 -0.04 (-0.14 to 0.06) 0.55
By categorization of breastfeeding 5 -0.02 (-0.09 to 0.05)
Ever breastfed 0.91
Breastfed for a given number of months 24 -0.01 (-0.07 to 0.06) 0.006
4 -0.20 (-0.33 to -0.06)
By control for confounding 28 -0.03 (-0.10 to 0.03)
None
Adjusted for socioeconomic and
demographic variables

By control for current measure of body size
No
Yes

Total

24 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Review 3 - Breastfeeding and the risk of overweight and obesity
in later life

It has been proposed that breastfeeding pro- not be considered as a major methodological
motion might be an effective way to prevent flaw in this meta-analysis. Existing cut-offs for
the development of obesity (102). overweight/obesity will have to be reassessed in
the light of the new WHO Growth Standards.
Biological plausibility
Mean body mass index or preva-
Several possible biological mechanisms for a lence of overweight/obesity
protective effect of breastfeeding against over-
weight and obesity have been proposed. Grummer-Strawn suggested that breastfeeding
may be associated both with a lower prevalence
Differences in protein intake and energy me- of overweight/obesity and with underweight in
tabolism may be one of the biological mecha- later life, due to a smaller variance of weight-
nisms linking breastfeeding to later obesity. related indices in subjects who were breastfed
Lower protein intake and reduced energy me- (32). Therefore, the mean body mass index
tabolism were reported among breastfed infants would remain unchanged but breastfeeding
(103). Rolland-Cachera et al (104) observed that would still have an effect on the upper tail of
higher protein intakes in early life, regardless of the body mass index distribution - that is, on
the type of feeding, was associated with an in- the prevalence of overweight/obesity.
creased risk of later obesity.
Overview of existing meta-
Another possibility is that breastfed and for- analyses
mula-fed infants have different hormonal re-
sponses to feeding, with formula feeding lead- The protective effect of breastfeeding against
ing to a greater insulin response resulting in fat childhood obesity was initially proposed by
deposition and increased number of adipocytes Kramer in 1981 (108). More recently, several
(105). studies were published on this topic. We iden-
tified four systematic reviews on the relation-
Finally, limited evidence suggests that ship between breastfeeding and overweight.
breastfed infants adapt more readily to new foods
such as vegetables, thus reducing the caloric In 2004, Arenz et al (107) were the first to
density of their subsequent diets (106). publish a systematic review of the evidence con-
cerning the protective effect of breastfeeding
Specific methodological issues duration against childhood obesity. To be in-
cluded in their meta-analysis, the studies had
General methodological issues affecting studies to fulfill the following criteria:
of the long-term consequences of breastfeeding
were addressed in the Introduction. In addition, z Analyses had to be adjusted for at least
the following methodological issues should be three of the following possible confo-
taken into account when studying overweight/ unders: birth weight, parental overweight,
obesity as the outcome. parental smoking, dietary factors, physi-
cal activity, and socioeconomic status.
Definition of overweight/obesity
z Odds ratios or relative risks had to be
Although different criteria and percentiles have reported.
been used in the definition of obesity, the re-
sults of the studies have been similar. Arenz et z Age at the last follow-up had to be be-
al (107) reported no difference in mean effect tween 5 and 18 years.
among studies using the 90th, 95th or 97th per-
centile to define obesity. Therefore, differences z Obesity had to be defined by body mass
in the definition of overweight/obesity should index percentiles >90, 95 or 97.

25

Only nine studies were included in this meta- tive effect of breastfeeding duration. This is a

analysis; 19 were not eligible. The main reasons strong suggestion of publication bias.

for exclusion were failure to report adequately Owen et al published two meta-analyses. In

adjusted estimates and a definition of obesity the first (109), the authors managed to obtain

that did not match the study criteria. Other odds ratio estimates from 28 of 61 studies re-

authors (see below) were more flexible in ac- porting on the relationship between

cepting different definitions of overweight, and breastfeeding and obesity. Unlike the Arenz

were able to include a larger number of studies meta-analysis (107), Owen et al (109) included

in their meta-analyses. studies that provided only crude odds ratios and

Analyses were stratified according to the fol- were flexible in terms of the definition of obes-

lowing study characteristics: type of design, age ity. Meta-regression analysis was used to inves-

group, definition of breastfeeding, number of tigate differences in the pooled odds ratio ac-

variables included in the multivariable analysis, cording to study size, age group at outcome

and definition of obesity. measurement, year of birth and attrition rate,

definition of obesity, and length of recall

Figure 3.1. Odds ratio of obesity according to study size, for information on breastfeeding duration.
modified from Owen et al. Figure 3.1 shows that small studies

tended to report stronger protective effects

Odds of obesity 1 of breastfeeding (pooled odds ratio: 0.43;
0.9 95% CI: 0.33–0.55) than larger studies
0.8 500 - 2500 > 2500 (pooled odds ratio: 0.88; 95% CI: 0.86–
0.7 0.90).
0.6
0.5 This protective effect may be due to
0.4 confounding by socioeconomic status and
0.3 parental body composition. In developed
0.2 countries, women who breastfeed tend to
have higher socioeconomic status and may
< 500 therefore be more “nutrition-conscious”
(21). Owen et al (109) identified 6 studies
Number of participants whose estimates were adjusted for the fol-

lowing confounders: socioeconomic status,

parental BMI, and maternal smoking. In

Their pooled odds ratio was 0.78 (95% CI: these studies, adjustment for confounding re-

0.71–0.85), and there was no sign of heteroge- duced the pooled odds ratio from 0.86 (95%

neity among the nine studies. The protective CI: 0.81–0.91) to 0.93 (95% CI: 0.88–0.99). This

effect of breastfeeding was independent of the suggests that at least part of the effect of

following study characteristics: study design (co- breastfeeding on body composition is due to

hort or cross-sectional); age at obesity assess- confounding by socioeconomic status and pa-

ment (<6 years or >6 years); definition of rental body composition, and that crude esti-

breastfeeding (never vs. ever or other definition); mates should not be included in the meta-analy-

and definition of obesity (95th or 97th percen- sis. Even after adjustment for socioeconomic

tile). On the other hand, the protective effect status, the possibility of residual confounding

of breastfeeding was slightly more pronounced cannot be ruled out.

in studies that adjusted their estimates for less Other methodological characteristics of the

than 7 variables (pooled odds ratio: 0.69; 95% studies, such as obesity definition and maternal

CI: 0.59–0.81), compared to those that adjusted recall of breastfeeding duration, were not re-

for 7 or more variables (pooled odds ratio: 0.78; lated to differences in the studies’ results.

95% CI: 0.70–0.87). As in the meta-analysis by Arenz et al (107),

The funnel plot was clearly asymmetric, with publication bias was evident with small studies

small studies tending to report a higher protec- reporting a stronger protective effect of

26 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

breastfeeding (Fig. 3.1). Selective reporting was the mean difference was greater among small

also evident; studies that failed to report odds studies (mean difference: -0.12; 95% CI: -0.29,

ratios were much less likely to conclude that 0.04) as compared to larger ones (mean differ-

breastfeeding was associated with a reduced risk ence: -0.03; 95% CI: -0.04, -0.01).

of obesity (1 of 35 studies), compared with stud- Despite the authors’ request, only 11 stud-

ies that did provide odds ratios (18 of 29 stud- ies provided estimates that were adjusted for

ies); this difference was statistically significant age, socioeconomic status, maternal smoking and

(P <0.001). However, because studies that did body mass index; among these studies, the crude

not report odds ratios were smaller than those mean difference of -0.12 (95% CI: -0.16, -0.08)

that presented such information, Owen et al disappeared in the adjusted analyses (mean dif-

(109) pointed out that their inclusion would ference: -0.01; 95% CI: -0.05, 0.03). This result

have a minimal impact on the pooled odds ra- reinforces the importance of controlling for

tio. confounding by socioeconomic status and ma-

The third meta-analysis was published by ternal body mass index when assessing the long-

Harder et al (110). Unlike the other analyses, term effect of breastfeeding duration on body

they attempted to assess the effect of duration composition.

of breastfeeding on the risk of overweight

(107,109), in search of a possible dose- Figure 3.2. Breastfeeding duration and odds ratio of over-
response association. Fourteen studies weight
providing results on more than one cat-
egory of breastfeeding duration were iden- 1.2

tified and included in the meta-analysis. Odds ratio of overweight 1
Subgroup analyses according to the defi- 0.8
nition of obesity and age at outcome as-

sessment were performed. 0.6
Fig. 3.2 shows that the odds ratio of 0.4

being overweight decreased continuously

with increasing duration of breastfeeding, 0.2
reaching a plateau after 9 months of
breastfeeding. Furthermore, trend analy- 0 1 to 3 4 to 6 7 to 9 >9
sis by a random-effect model showed that <1
each month of increase in breastfeeding
Duration of breast feeding (months)

duration was associated with a 4% de-

crease in the odds of overweight (OR: Update of existing meta-analyses
0.96; 95% CI: 0.94–0.98).

Because previously published meta-analyses A new meta-analysis was carried out including
(107,109,110) treated overweight/obesity as the recently published studies described above,
dichotomous variables, Owen et al (111) con- all the papers included in previously published
ducted a fourth meta-analysis to assess the ef- meta-analyses, and those identified by the two
fect of breastfeeding on mean BMI. The authors independent literature searches at WHO and
of the 70 identified studies were asked to pro- at the University of Pelotas. It was possible to
vide information on mean differences in BMI, include 33 studies with 39 estimates on the ef-
according to breastfeeding duration; these esti- fect of breastfeeding on prevalence of over-
mates were adjusted for age, socioeconomic sta- weight/obesity (Table 3.1). The forest plot shows
tus, maternal BMI, and maternal smoking in that results were clearly heterogeneous (Fig. 3.3).
pregnancy. In a random-effects model, including all stud-
ies, breastfed individuals were less likely to be
In the fixed-effects model including 36 stud- considered as overweight/obese, and the pooled
ies, breastfed subjects had lower mean BMI ratio was 0.78 (95% CI: 0.72–0.84).
(mean difference: -0.04; 95% CI: -0.05, -0.02).
In spite of efforts to prevent publication bias,

RESULTS AND DISCUSSION - REVIEW 3 27

Table 3.1. Breastfeeding and overweight/obesity in later life: studies included in the meta-analysis in ascending order of subjects' age at which outcome was measured

28 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES Author, year (reference) Study design Year of birth Age measured Gender Comparison groups Outcome Odds ratio (95%
of subjects confidence interval)
Obesity 0.84 (0.41–1.72)
Strbak 1991 (114) Cohort Not stated 1-7 years All Breastfed for >1 months (n=741) vs. Breastfed <2 weeks
(n=165) Obesity 0.6 (0.3–1.3)
1.18 (0.91–1.54)
Li 2005 (115) Cohort 1990-4 2-6 years All Breastfed for ≥4 months vs. Never breastfed (n=923)* Obesity 0.86 (0.44–1.65)
He 2000 (116) Case-control
Poulton 2001 (117) Cohort 1989-93 3-6 years All Breastfeeding vs. no breastfeeding (n=465)* Overweight or 0.72 (0.65–0.79)
obesity
1972-3 3 years All Breastfed for >6 months vs. Never breastfed (n=677)* 0.63 (0.41–0.96)
Obesity 0.84 (0.62–1.13)
Armstrong 2002 (118) Cohort 1995-6 3 years All Exclusively breastfed (n=8751) vs. Bottle fed at 6-8 0.72 (0.65–0.80)
weeks (n=23449) Overweight only
Obesity 1.0 (0.7–1.5)
Hediger 2001 (119) Cross-sectional 1983-91 3-5 years All Ever breastfed (n=1158) vs. Never breastfed (n=1498) 1.83 (0.53–6.28)
Overweight or
Grummer-Strawn 2004 Cohort 1988-92 4 years All Breastfed for ≥12 months (n=293) vs. Never breastfed obesity 0.61 (0.28–1.32)
(32) Cohort (n=7084)
1998 4 years All Obesity 0.91 (0.69–1.21)
Dubois 2006 (120) Breastfed for ≥3 months (n=710) vs. Breastfed for <3 0.64 (0.24–1.70)
All months (n=740) Overweight or
Araujo 2006 (121) Cohort 1993 4 years All obesity 0.71 (0.43–1.25)
Li 2003 (122) 4-8 years Ever breastfed (n=1221) vs. Never breastfed (n=52) 0.66 (0.25–1.78)
Cross-sectional 1983-87 All Obesity
All Breastfed for >9 months (n=194) vs. Breastfed <1 week 0.79 (0.68–0.93)
Maffeis 1994 (A) (123) Cross-sectional Not stated 4-12 years (n=505) Obesity 0.75 (0.57–0.98)
Scaglioni 2000 (124) 5 years 0.90 (0.70–1.10)
Cohort 1991 Not stated Overweight or 0.60 (0.40–0.90)
obesity
Ever breastfed (n=124) vs. Never breastfed (n=23)
Obesity
O'Callaghan 1997 (125) Cohort 1981-4 5 years All Ever breastfed (n=3119) vs. Never breastfed (n=790)
Overweight or
Burdette 2006 (126) Cohort Not stated 5 years All Ever breastfed (n=231) vs. Never breastfed (n=82) obesity

Von Kries 1999 (127) Cross-sectional 1991-2 5-6 years All Ever breastfed (n=5184) vs. Never breastfed (4022) Overweight or
obesity
Frye 2003 (128) Cross-sectional Not stated 5-14 years All Exclusively breastfed for >12 weeks (n=762) vs. Never Obesity
* total sample breastfed (n=1955)
Overweight or
obesity
Obesity

Table 3.1. (continued) Year of birth Age measured Gender Comparison groups Outcome Odds ratio (95%
Author, year (reference) Study design of subjects confidence interval)
Wadsworth 1999 (129) Cohort Overweight only
1946 6 years All Ever breastfed (n=2873) vs. Never breastfed (858) Obesity 0.94 (0.73–1.20)
0.88 (0.59–1.32)
Thorsdottir 2003 (130) Cohort Not stated 6 years Male Breastfed for ≥6 months vs. Breastfed for <6 months Overweight or
All (n=61)* obesity 0.33 (0.13–0.83)
Bergmann 2003 (131) Cohort 1990 6 years All
All Breastfed for ≥3 months (n=607) vs. Breastfed for <3 Overweight only 0.53 (0.31–0.89)
Reilly 2005 (132) Cohort 1991-2 7-8 years All months (n=280) Obesity 0.46 (0.23–0.92)

Exclusively breastfed for ≥2 months (n=2211) vs. Never Obesity 1.22 (0.87–1.71)
breastfed (n=1153)
Eid 1970 (133) Cohort 1961 8 years Obesity 0.42 (0.11–1.61)
Liese 2001 (134) 9-10 years Not stated 0.66 (0.52–0.87)
Cross-sectional 1985-7 Overweight or
Ever breastfed (n=1754) vs. Never breastfed (n=354) obesity 0.45 (0.27–0.78)

Sung 2003 (A) (135) Cross-sectional 1990 9-12 years All Not stated Overweight or 0.80 (0.71–0.90)
obesity 0.80 (0.66–0.96)
Toschke 2002 (136) Cross-sectional 1977-85 6-14 years All Ever breastfed (n=30641) vs. Never breastfed (n=3127) 0.36 (0.1–1.28)
Overweight or 0.73 (0.23–2.27)
Poulton 2001 (117) Cohort 1972-3 11 years All Breastfed for >6 months vs. Never breastfed (n=591)* obesity
Li 2003 (122) 9-18 years All Obesity
Cross-sectional 1973-82 Breastfed for >9 months (n=67) vs. Breastfed <1 week
Gillman 2001 (137) (n=471) Overweight

Li 2005 (115) Mostly or only breastfed (n=4510) vs. Mostly or only Obesity
Elliott 1997 (138) formula fed (n=2248)
RESULTS AND DISCUSSION - REVIEW 3 Cross-sectional Not stated 9-14 years All Overweight only 0.95 (0.84–1.07)
Kramer 1981 (108) Breastfed for ≥4 months vs. Never breastfed (n=749)* Obesity 0.78 (0.66–0.91)
Kramer 1981 (108) Cohort 1982-5 1-14 years All
Tulldahl 1999 (139) Cohort 1977-80 12-17 years All Breastfed for >2 months (n=73) vs. Breastfed for ≤2 Obesity 0.6 (0.3–1.6)
months (n=63)
* total sample Overweight 0.51 (0.22–1.16)
Ever breastfed (n=45) vs. Never breastfed (284)
Case-control Not stated 12-18 years All Obesity 0.44 (0.21–0.93)
Case-control Not stated 12-18 years All Ever breastfed (n=60) vs. Never breastfed (332) 0.29 (0.11–0.73)
Cohort 1979 15-16 years All Obesity 0.66 (0.44–0.98)
Breastfed for >2 months (n=390) vs. Breastfed for ≤2
29 months (n=391) Overweight or
obesity

Odds ratio (95% Table 3.2 shows that there was no marked
confidence interval) effect modification by age group, year at birth,
0.73 (0.50–1.07) control of confounding, categories of breastfeed-
0.79 (0.46–1.34) ing duration, study setting, study design, and
0.64 (0.33–1.26) control for confounding variables. On the other
0.34 (0.12–1.01) hand, the study’s outcome was related to het-
0.93 (0.74–1.17) erogeneity between studies.
0.84 (0.67–1.05)
0.73 (0.51–1.07) Concerning publication bias, the funnel plot
1.10 (0.88–1.37) is quite asymmetrical, with small studies tend-
ing to report a higher protective effect of
Outcome breastfeeding. Indeed, Table 3.2 shows that the
Obesity protective effect of breastfeeding was higher
Overweight among small studies (<500 participants). How-
Overweight or ever, studies with 500-1500 and >1500 partici-
obesity pants had similar protective effects, which were
Obesity also similar to the pooled protective effect of
Obesity all studies.
Obesity
Obesity Eight studies provided odds ratios for more
Obesity than one outcome, such as overweight only (e.g.
BMI 25-29.9), overweight plus obesity (e.g. BMI
Year of birth Age measured Gender Comparison groups >25) and obesity (e.g. BMI >30). Six of theseMales Breastfed for ≥1 month (n=489) vs. Breastfed for <1Breastfed for >6 months vs. Never breastfed (n=687)*Breastfed for ≥4 months (n=121) vs. Never breastfedMales Breastfed for >1 month vs. Never breastfed (n=4662) *Ever breastfed (n=1999) vs. Never breastfed (n=386)Breastfed for >8 months (n=986) vs. Breastfed for <2
of subjects eight studies reported a more marked protec-month (n=1666) (n=78) Females Breastfed for >1 month vs. Never breastfed (n=4625) * months (n=607)
tive effect against obesity than against over-
weight only or overweight plus obesity. This sup-All 31-35 years All All All
ports a causal effect of breastfeeding.
18 years 21 years 33 years
Relevant studies not included in Adult life
the meta-analyses 56-66 years

A frequent limitation of observational studies19821972-3 1966 1958
is inadequate adjustment for confounding by 1957-9
socioeconomic and maternal variables. In the 1934-44
absence of randomized studies, within-family
Table 3.1. (continued) analyses allow controlling for confounding byCohortCohort Cohort
Author, year (reference) Study design socioeconomic, maternal variables, as well as self-Cohort Cohort
selection bias. Gillman et al (112) analysed data Cohort
from 5614 sibling sets from the Growing Up
Today Study to assess the association ofVictora 2003 (140)Poulton 2001 (117) Parsons 2003 (142) * total sample
breastfeeding with adolescent obesity withinKvaavik 2005 (141) Richter 1981 (143)
sibling sets. The overall odds of overweight in Eriksson 2003 (144)
that study were smaller among subjects breastfed
for at least 7 months, compared with those
breastfed for 3 months or less (OR: 0.85; 95%
CI: 0.71–1.00). When the analyses were limited
to the 172 families in which one sibling was
breastfed for at least 7 months and another for
3 months or less, the resulting odds ratio was
similar to that of the overall analyses OR: 0.89;
95% CI: 0.50–1.59), but the upper 95% confi-
dence limit was well above the unity (1). This
similarity between the results obtained in the

30 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Figure 3.3. Odds ratio and its 95% confidence interval of being considered as overweight/obese, compar-
ing breastfed vs. non-breastfed subjects in different studies. Whether the estimate was for
males (M), females (F) and all (A) is indicated in parenthesis

Strbak 1991 (A) 0.5 1 25 10
He 2000 (A) Favours breastfeeding Favours not breastfeeding

Poulton 2001 (A)
Armstrong 2002 (A)

Hediger 2001 (A)
Grummer-Strawn 2004 (A)

Li 2003 (A)
Li 2003 (A)
Scaglioni 2000 (A)
O'Callaghan 1997 (A)
Von Kries 1999 (A)
Frye 2003 (A)
Wadsworth 1999 (A)
Bergmann 2003 (A)
Li 2005 (A)
Li 2005 (A)
Dubois 2006 (A)
Araujo 2006 (A)
Maffeis 1994 (A)
Burdette 2006 (A)
Reilly 2005 (A)
Eid 1970 (A)
Liese 2001 (A)
Sung 2003 (A)
Toschke 2002 (A)
Poulton 2001 (A)
Gillman 2001 (A)
Elliott 1997 (A)
Kramer 1981 (A)
Kramer 1981 (A)
Tulldahl 1999 (A)
Victora 2003 (A)
Poulton 2001 (A)
Kvaavik 2005 (A)
Parsons 2003 (M)
Parsons 2003 (F)
Richter 1981 (A)
Eriksson 2003 (A)
Thorsdottir 2003 (M)

Combined

0.1

Odds ratio of overweight/obesit

total sample and those from the siblings study size for the within-family analysis was quite small
suggests that confounding by socioeconomic leading to less precise estimates. Another within-
status was not an important issue in this study. family analysis from the United States (113)
However, because heterogeneity in breastfeeding found no association between breastfeeding
durations among siblings is much smaller than duration and prevalence of obesity.
for unrelated individuals, the effective sample

31RESULTS AND DISCUSSION - REVIEW 3

Table 3.2. Breastfeeding and the risk of overweight and obesity in later life: Random-effects meta-
analyses of risk of overweight/obesity by subgroup

Subgroup analysis Number of estimates Pooled odds ratio and P value
95% confidence interval
By age group 22 0.001
1 to 9 years 11 0.79 (0.71 to 0.87) 0.001
9 to 19 years 6 0.69 (0.60 to 0.80)
>19 years 0.88 (0.74 to 1.04) 0.13
11
By study size 11 0.51 (0.35 to 0.75) 0.001
<500 participants 17 0.79 (0.66 to 0.93) 0.006
500–1499 participants 0.80 (0.74 to 0.87) 0.001
≥1500 participants 13
22 0.83 (0.73 to 0.95) 0.008
By year at birth 0.78 (0.72 to 0.85) 0.001
Before 1980 26
After 1980 3 0.79 (0.72 to 0.87) 0.001
10 0.58 (0.23 to 1.45) 0.24
By study design 0.75 (0.69 to 0.83)
Cross-sectional 24 0.001
Case-control 15 0.79 (0.71 to 0.87)
Cohort 0.76 (0.67 to 0.86) 0.001
12 0.001
By length of recall of breastfeeding 23 0.75 (0.67 to 0.83)
<3 years 0.78 (0.71 to 0.86) 0.001
≥3 years 16 0.001
3 0.76 (0.64 to 0.91)
By categorization of breastfeeding 20 0.72 (0.66 to 0.79) 0.004
Ever breastfed 0.77 (0.71 to 0.84) 0.001
Breastfed for a given number of months 33 0.001
6 0.77 (0.71 to 0.83)
By control for confounding 39 0.82 (0.62 to 1.09) 0.001
None 0.78 (0.72 to 0.84) 0.18
Adjusted for socioeconomic status
Adjusted for socioeconomic status and
parental anthropometry

By study setting
High-income country
Middle/Low-income country

Total

Table 3.3. Comparison of odds ratios for overweight only*, overweight plus obesity* and for obes-
ity* only, in studies reporting more than one of these outcome

Study Overweight only Overweight plus obesity Obesity
Hediger 2001 (119) 0.63 - 0.84
Von Kries 1999 (127) -
Frye 2003(128) - 0.79 0.75
Wadsworth 1999 (129) 0.94 0.90 0.60
Bergmann 2003 (131) 0.53
Toschke 2002 (136) - - 0.88
Gillman 2001 (137) 0.95 - 0.46
Kvaavik 2005 (141) - 0.80 0.80
- 0.78
0.64 0.34

* For example in adult subjects, overweight only defined as BMI 25-29.9, overweight plus obesity defined as
BMI ≥25 and obesity only defined as BMI ≥30 kg/m2

32 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Conclusion Figure 3.4. Funnel plot showing odds ratio for overweight/obes-
ity by standard error of odds ratio
The evidence suggests that breastfeed-

ing may have a small protective effect

on the prevalence of obesity. In spite 0
of the evidence of publication bias, a

protective effect of breastfeeding was 0.2
still observed among the larger studies

(>1500 participants), suggesting that Standard error

this association was not due to publi- 0.4
cation bias. With respect to confound-

ing, studies that controlled for socio- 0.6
economic status and parental anthro-

pometry also reported that breastfeed-

ing was associated with a lower preva- 0.8
lence of obesity. This effect seems to
be more important against obesity 1 23
than against overweight. Odds ratio (log scale)

Because the great majority of the

published studies were conducted in

Western Europe and North America, we are not

able to assess whether this association is present

in low and middle-income settings.

RESULTS AND DISCUSSION - REVIEW 3 33

Review 4 - Breastfeeding and the risk of type-2 diabetes

Biological plausibility collected from the medical records. The odds
ratio for type-2 diabetes was 0.51 (95% CI: 0.3–
Two possible mechanisms for a protective ef- 0.9), comparing exclusively breastfed subjects
fect of breastfeeding against type-2 diabetes have with those who were partially or exclusively
been proposed. bottle-fed.

Baur et al (145) observed that the fasting Young et al (150) assessed the role of prena-
glucose level was inversely correlated to long- tal and early infancy risk factors for type-2 dia-
chain polyunsaturated fatty acids in skeletal betes among native Canadians. Forty-six cases
muscle membranes. Because long-chain polyun- were recruited from the only clinical centre for
saturated fatty acids are present in breastmilk, the treatment of diabetes in the province of
but not in most brands of formula (49), it has Manitoba. Two controls were chosen for each
been postulated that changes in skeletal muscle case, from a paediatric clinic. Mean ages were
membrane would play a role in the development 14.0 years for the cases and 12.7 years for con-
of insulin resistance, leading to compensatory trols. The odds ratio for type-2 diabetes between
hyperinsulinaemia. Over a period of time there subjects who were breastfed for >6 months and
would be β-cell failure, resulting in the occur- those breastfed for <6 months was 0.36 (95%
rence of type-2 diabetes (146). CI: 0.13–0.99).

Several studies (105,147,148) reported that Martin et al (58) studied a cohort of sub-
formula-fed infants have higher basal and post- jects recruited in 16 centres in England and Scot-
prandial concentrations of insulin and land, who participated in a 1-week survey of
neurotensin, which modulates insulin and glu- diet and health when aged <20 years between
cagon release. These differences may lead to the 1937 and 1939. Information on breastfeeding
earlier development of insulin resistance and duration was obtained from the subjects’ moth-
type-2 diabetes. ers. At a mean age of 71 years, breastfeeding
was not associated with type-2 diabetes (odds
These are possible mechanisms for an asso- ratio: 0.97; 95% CI: 0.41–2.30).
ciation between breastfeeding and diabetes.
Other biological mechanisms may also exist, Rich-Edwards et al (151) studied a cohort
which are presently unknown. of registered nurses in the US, who had been
followed since 1976. Information on occurrence
Overview of the evidence of diabetes was reported by the subjects them-
selves. Prevalence of diabetes was lower among
The two independent literature searches carried breastfed (4%) compared to non-breastfed (5%)
out at WHO and at the University of Pelotas subjects.
identified five papers that assessed the associa-
tion between breastfeeding duration and type-2 These five studies were included in a meta-
diabetes. analysis. The pooled odds ratio was 0.63 (95%
CI: 0.45–0.89). Fig. 4.1 shows the forest plot
Pettitt et al (149) followed a cohort of Pima for this analysis.
Indians born between 1950 and 1978. Informa-
tion on infant feeding was provided by mothers Other relevant studies not in-
in 1978. The odds ratio for type-2 diabetes be- cluded in the meta-analysis
tween subjects who were exclusively breastfed
in the first two months and those who were Lawlor et al (60) evaluated a random sample of
exclusively bottle-fed was 0.41 (95% CI: 0.18– schoolchildren aged 9 and 15 years from Esto-
0.93). nia and Denmark. Insulin resistance was esti-
mated according to the homeostasis model as-
Ravelli et al (77) studied a cohort of sub- sessment (HOMA), based on fasting glucose and
jects who were born between 1943 and 1947 in insulin levels. Information on infant feeding was
a university hospital in Amsterdam. Informa- collected from the parents at the time of the
tion on infant feeding at hospital discharge was

34 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

subjects’ examination. After controlling for pos- who were born pre-term and randomized to
sible confounding variables, there was no asso- receive a nutrient-enriched or lower-nutrient
ciation between breastfeeding and insulin re- diet. The level of 32-33 split proinsulin was lower
sistance. Difference in HOMA index was -4% among those subjects who were randomized to
(95% CI: -13 to 5) between ever and never receive banked breastmilk, compared to those
breastfed subjects. Also, there was no evidence receiving pre-term formula; the difference was
of a dose-response trend. not statistically significant (P = 0.07).

Martin et al (57) also studied a cohort of Conclusion
men from Caerphilly, South Wales, aged 45-59
years when examined between 1979 and 1983. Evidence on a possible programming effect of
Information on breastfeeding duration was ob- breastfeeding on glucose metabolism is sparse.
tained from the subjects’ mothers or a close fe- Studies assessing the risk of type-2 diabetes re-
male relative. Difference in the HOMA index ported a protective effect of breastfeeding, with
between breastfed and bottle-fed subjects was a pooled odds ratio of 0.63 (95% CI: 0.45–0.89)
0.00 (95% CI: -0.10 to 0.10). in breastfed compared to non-breastfed subjects.
On the other hand, two other studies failed to
Plancoulaine et al (94) evaluated a sample report an association between HOMA index, a
of children aged 5 to 11 years in two small towns measure of insulin resistance, and breastfeeding
in northern France. Information on breastfeed- duration, and a study on fasting blood glucose
ing duration was obtained from the mother at levels was also negative. At this stage, it is not
the moment of the children’s examination. Fast- possible to draw firm conclusions about the long-
ing blood glucose levels were similar among term effect of breastfeeding on the risk of type-
breastfed and non-breastfed children. 2 diabetes and related outcomes. Further stud-
ies are badly needed on this topic.
Singhal et al (152) assessed 32-33 split
proinsulin concentration – a marker of insulin
resistance – among subjects aged 13-16 years

Figure 4.1. Odds ratio and 95% confidence interval of having type-2 diabetes in different studies, compar-
ing breastfed vs. non-breastfed subjects. Whether the estimate was for males (M), females (F)
and all (A) is indicated in parenthesis.

Young 2002 (A)
Petit 1997 (A)

Ravelli 2000 (A)
Martin 2005 (A)
Rich-Edwards 2004 (A)

Combined 0.5 1 2
0.1
Favours breastfeeding Favours not breastfeeding
Odds ratio of type-2 diabetes

RESULTS AND DISCUSSION - REVIEW 4 35

Review 5 - Breastfeeding and school achievement/intelligence
levels

Biological plausibility of stimulation. In addition, in societies where
breastfeeding is more common among upper
Long-chain polyunsaturated fatty acids are social groups, the possibility of confounding by
present in breastmilk, but not in most brands parental education level has to be addressed.
of formula (49). These fatty acids are preferen-
tially incorporated into neural cell membranes; Overview of the evidence
structural lipids constitute about 60% of the
human brain. The major lipid components in- We identified three meta-analyses that assessed
clude docosahexaenoic (DHA) and arachidonic the relationship between breastfeeding duration
(AA) acids (153), which are important for reti- and performance in intelligence tests.
nal and cortical brain development (154). Bjerve
et al (155) reported that the results of the Bayley In 1999, Anderson et al (163) were the first
mental and psychomotor development indexes to publish a meta-analysis on this topic, em-
correlated positively with serum DHA concen- ploying the following selection criteria:
trations.
z the study included a comparison between
AA and DHA accumulate in the brain and subjects who were mostly breastfed with
retina most rapidly during the last trimester of those who were mostly bottle-fed;
pregnancy and the first months after birth (156).
Their reserves are limited at birth, especially in z the outcome was measured with a widely
pre-term infants, and decline rapidly when lack- applied test of cognitive development or
ing in the diet (157). Bottle-fed infants have ability, yielding a single score;
been shown to have lower long-chain polyun-
saturated fatty acids in the phospholipids of the z subjects were examined between infancy
cerebral cortex than infants who are fed and adolescence.
breastmilk (158). This is, therefore, a potential
mechanism for an effect of breastfeeding on in- The studies were considered as including con-
tellectual development. trol for confounding if the estimates were ad-
justed for a minimum of five variables, from a
In addition to the chemical properties of list of 15 potential confounders.
breastmilk, breastfeeding enhances the bonding
between mother and child (159, 160), which may Eleven studies were included in their analy-
contribute to the child’s intellectual develop- ses. In a random-effects model, the adjusted
ment. mean difference in cognitive function was of
3.16 (95% CI: 2.35 to 3.98) points in favour of
Specific methodological issues breastfed subjects. The effect of breastfeeding
was not modified by age at measurement. The
General methodological issues were addressed benefit of breastfeeding was higher among low
in the Introduction. Two points deserve special birthweight infants (mean difference: 5.18
attention. Because cognition and performance points; 95% CI: 3.59 to 6.77) although a sig-
in intelligence tests are positively related to the nificant effect was also observed among normal
stimulation received by the child (161) and be- birthweight subjects (mean difference: 2.66
cause breastfeeding mothers may be more prone points; 95% CI: 2.15 to 3.17).
to stimulating their children (162), studies as-
sessing the long-term consequences of In 2000, Drane et al (164) carried out a sec-
breastfeeding on intellectual performance should ond systematic review. Articles had to fulfil the
attempt to control for the quantity and quality following criteria:

z subjects had to be born between 1960
and 1998;

z cognition had to be measured by using
standard tests;

36 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

z breastfeeding had to be measured as a ies controlled for both socioeconomic status and
categorical variable (exclusively breastfed, stimulation/interaction of the child but did not
partially breastfed, exclusively formula- fulfil the remaining methodological criteria.
fed) or as a “continuous variable” (dura- Three of these studies reported that perform-
tion of exclusive breastfeeding or propor- ance in the intelligence test was higher among
tion of the diet as breastmilk); breastfed subjects (165,174,175), whereas four
failed to show an association (162,176-178).
z analyses had to be adjusted for socioeco- Given the conflicting results from high-quality
nomic status and birth weight, at least. studies, the authors stated that the evidence on
the effect of breastfeeding on cognition was not
Twenty-four studies including subjects born convincing.
between 1960 and 1998 were identified, but
only five (165-169) fulfilled the three methodo- A very important study that was included
logical pre-requisites. Another study (170), in the meta-analyses carried out by Anderson
which met two standards (confounding and cog- et al (163) and by Drane et al (164) was a
nition assessment) and partially met the stand- randomized trial by Lucas et al (167), in which
ard for breastfeeding measurement, was also newborn pre-term babies received breastmilk
included. Four of these studies reported a posi- or formula, and a significant improvement in
tive effect of breastfeeding on cognitive devel- WISC-R was found in the former. Because of
opment, particularly for low-birthweight sub- the randomized design, these differences are not
jects. likely to be due to confounding.

In 2002, Jain et al (171) published a third Update of existing meta-analysis
review. Inclusion was restricted to studies pub-
lished in English, and each study was assessed Jain and colleagues (171) restricted their inter-
for the following methodological aspects: study pretation to whether or not the study results
design, sample size, target population, quality had been significant. They did not discuss the
of feeding data, control of susceptibility bias, direction of the effect in non-significant stud-
blinding and outcome measures. Of the 40 stud- ies, nor did they try to pool the results in a
ies identified, only nine met all criteria for qual- meta-analysis. We carried out a meta-analysis
ity of the feeding data; the breastfed group was including these papers, as well as three more
specified as those who mostly breastfed, infor- recent articles identified in our own systematic
mation on infant feeding was collected in in- review. According to the criteria proposed by
fancy and by interviewing the mother or from Jain and colleagues, all estimates were adjusted
health records, and duration of breastfeeding for stimulation at home and fulfilled the other
was of at least 1 month among the breastfed quality criteria. Studies restricted to very low
subjects. Only two of these nine studies pre- birthweight infants were not included. The three
sented estimates adjusted for socioeconomic additional studies are described below.
status and stimulation at home, and their re-
sults were conflicting. Wigg et al (172) reported Quinn et al (179) studied a cohort of 7357
that, after controlling for socioeconomic status singleton children whose mothers had been en-
and quality of the child’s environment, breastfed rolled in the Mater Hospital-University of
subjects presented a small advantage in the Queensland Study of Pregnancy. Information on
Wechsler Intelligence Scale for Children at breastfeeding duration was obtained from the
11-13 years of age (mean difference: 0.8; 95% mother when the child was six months old. At
CI: -1.9 to 3.5 points). Johnson et al (173) re- five years, 4049 children were assessed with the
ported that the adjusted difference between Peabody Picture Vocabular y Test Revised
breastfed and non-breastfed 3-year-olds in the (PPVT-R). Breastfeeding duration was positively
Stanford-Binet Composite IQ Scale was 5.0 associated with the PPVT-R score, and after con-
points (95% CI: 0.3 to 9.5), while in the Peabody trolling for confounding by socioeconomic sta-
Picture Vocabulary Test the difference was 4.6 tus, birthweight, and stimulation in the home,
points (95% CI: 0.7 to 8.5). Seven other stud- the mean score for children breastfed for six

RESULTS AND DISCUSSION - REVIEW 5 37

months or more was 8.2 (95% CI: 6.5 to 9.9) with academic performance in high-school leav-
points higher for females and 5.8 (95% CI: 4.1 ing examinations. A lower percentage of chil-
to 7.5) points higher for males, when compared dren left high school without qualifications in
to those never breastfed. the group breastfed for >8 months (14%) as
compared to those breastfed for 4–7 months
Clark et al (180) followed a cohort of chil- (16.4%), <4 months (19.2%), and not breastfed
dren born between 1991 and 1996 in urban com- (22.2%) (overall P <0.05).
munities near Santiago, Chile, who were enrolled
in a study on prevention of iron deficiency in Richards et al (182) reported from the Brit-
healthy full-term infants. At six months, the in- ish 1946 birth cohort that the odds ratio for
fants were randomly assigned to receive iron obtaining an advanced educational qualification
supplementation. At five years, information from by the age of 26 years was directly associated
784 (62.6%) subjects was gathered. Children with breastfeeding duration, showing a dose-
breastfed for <2 months or >8 months had lower response relationship. The odds of obtaining an
scores for language, motor and cognition tests advanced educational qualification were 1.58
than those breastfed for 2-8 months, after ad- (95% CI: 1.15–2.18) times higher among those
justment for socioeconomic factors and home subjects who were breastfed for more than seven
stimulation. months, compared with those who were never
breastfed.
Angelsen et al (181) reported from Norway
that the odds of having a low total intelligence Breastfeeding was also associated with in-
quotient at five years of age was higher among creased achieved schooling among Brazilian ado-
those children breastfed for <3 months com- lescents who had been followed-up since birth
pared to those breastfed for at least 6 months (183). In the adjusted analyses, there was an
(odds ratio: 1.5; 95% CI: 1.0–2.1). increase in schooling with breastfeeding dura-
tion up to 12 months. Subjects breastfed for 9-
Because the Chilean and Norwegian studies 11 months had achieved 8.0 (95% CI: 7.5–8.5)
did not provide results in terms of means scores, years of schooling, while those breastfed for less
they were not included in the meta-analyses. Fig. than one month had 7.2 (95% CI: 6.9–7.6) years.
5.1 shows that the eight studies observed a ben- Children breastfed for 12 or more months at-
eficial effect from breastfeeding, which was sta- tained on average 7.7 (95% CI: 7.3–8.0) years.
tistically significant in six of them. Because het- This association is unlikely to be explained by
erogeneity among studies was statistically sig- residual confounding by socioeconomic status.
nificant, a random-effects model was used. In In this population, there was no strong associa-
the pooled analysis, performance in intelligence tion between breastfeeding and social class, and
tests was higher among those subjects who had the poorest children were those breastfed for
been breastfed (mean difference: 4.9; 95% CI: over one year. In addition, a stratified analysis
2.97 to 6.92). Given the small number of stud- showed that breastfeeding was associated with
ies, it was not possible to assess publication bias. increased schooling within all strata of family
income.
Studies on breastfeeding and
schooling Conclusion

Our systematic review resulted in only three This meta-analysis suggests that breastfeeding
studies on the relationship between breastfeed- is associated with increased cognitive develop-
ing and achieved schooling. Horwood & ment in childhood, in studies that controlled
Fergusson (168) studied adolescents from for confounding by socioeconomic status and
Christchurch, New Zealand, who had been fol- stimulation at home. The practical implications
lowed since birth. Even after controlling for con- of a relatively small increase in the performance
founding by socioeconomic status and perina- in developmental tests in childhood may be
tal variables (birthweight, birth order, and ma- open to debate. However, evidence from the only
ternal smoking in pregnancy), increased three studies on school performance in late ado-
breastfeeding duration was positively associated

38 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

lescence or young adulthood suggests that contributes to intellectual development. Al-
breastfeeding is also positively associated with though in observational studies it is not possi-
educational attainment (168,182,183). ble to disentangle these two effects, the posi-
tive results from the randomized trial carried
The issue remains of whether the associa- out by Lucas et al (167) suggest that the nutri-
tion is related to the properties of breastmilk tional properties of breastmilk alone seem to
itself, or whether breastfeeding enhances the have an effect.
bonding between mother and child, and thus

Figure 5.1. Mean difference in cognitive development scores and its 95% confidence interval between
breastfed and non-breastfed subjects in different studies. Whether the estimate was for males
(M), females (F) and all (A) is indicated in parenthesis

Johnson (A) 1996
Morrow-Tlucak (A) 1988

Rogers (A) 1978
Jacobson (A) 1992

Lucas (A) 1992
Wigg (A) 1998
Quinn (M) 2001
Quinn (F) 2001

Combined

-2 -1 0 1 2 3 4 5 6 7 8 9 10 15
Mean difference
Mean lower in breastfed Mean higher in breastfed
subjects subjects

RESULTS AND DISCUSSION - REVIEW 5 39

VI. Conclusions

The available evidence suggests that breastfeed- fect was similar to that of dietary advice in
ing may have long-term benefits. Subjects who adulthood. Similarly, for the prevention of type-
had been breastfed were found to have a lower 2 diabetes, the magnitude was similar to that
mean blood pressure and lower total cholesterol, of diet and physical activity. Concerning obes-
and showed higher performance in intelligence ity, whereas Summerbell et al (184) reported
tests. Furthermore, the prevalence of overweight/ that combined dietary education and physical
obesity and type-2 diabetes was lower among activity interventions were not effective in re-
breastfed subjects. All effects were statistically ducing childhood obesity and overweight, we
significant, but for some outcomes their mag- noticed that breastfeeding was associated with
nitude was relatively modest. a 22% reduction in the prevalence of overweight/
obesity.
The Table 6.1 summarizes the magnitude of
the effects of breastfeeding based on the five This Table is intended for illustrative pur-
meta-analyses described above. For all outcomes, poses only. It should be interpreted with cau-
except performance in intelligence tests, we pro- tion because it includes a comparison of the
vide a comparison of these effects with those effect of actual interventions – none of them
observed for other public health interventions. with perfect compliance levels – with the gross
For blood pressure, the effect of breastfeeding difference of the effect between breastfed and
was smaller than those derived from other pub- non-breastfed subjects, which corresponds to an
lic health interventions targeted at adults, such intervention with 100% compliance. Only the
as dietary advice, physical activity, salt restric- long-term follow-up of subjects involved in
tion, and multiple risk factor interventions. On breastfeeding trials will provide a more accu-
the other hand, for total cholesterol among rate estimate of the impact of breastfeeding
adults, the magnitude of the breastfeeding ef- promotion.

40 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

Table 6.1. Summary of effect of breastfeeding and comparison with other interventions

Outcome Range of effect of other public health interventions in later life Breastfeeding
(current review)

Diet / Dietary advice Exercise Multiple risk factor Modest salt restriction Pooled effect size (95% Conclusion
intervention confidence interval)

Blood pressure The effect of breastfeeding is
(mean difference in mm significant, but smaller than the effect
Hg, 95% CI) of other interventions.

Systolic -2.1 (-2.8 to -1.4) -3.84 (-4.97 to -2.72) -4.2 (-4.6 to -3.8) -2.03 (-2.56 to -1.50) -1.21 (-1.72 to -0.70)
Diastolic -1.6 (-2.7 to -0.6) -2.58 (-3.35 to -1.81) -2.7 (-2.9 to -2.5) -0.99 (-1.4 to -0.57) -0.49 (-0.87 to -0.11)

ref. (45) ref. (185) ref. (186) ref. (187)

Total serum cholesterol -0.13 (-0.23 to -0.03) -0.14 (-0.16 to -0.12) -0.18 (-0.30 to -0.06) The effect of breastfeeding is
(mean difference in ref. (45) ref. (186) significant and larger than the effect
mmol/L, 95% CI) Odds ratio of other interventions.
0.78 (0.72 to 0.84)
Overweight or obesity Diet/dietary advice and exercise The effect of breastfeeding is
5/6 studies showed no effect on childhood significant (22% reduction), while
obesity. Meta-analysis not done because of other interventions showed no effect.

heterogeneity between studies.

ref. (184)

Type II diabetes Diet/dietary advice and exercise
31-46% reduction in risk in persons with
Odds ratio The effect of breastfeeding is
impaired glucose tolerance. 0.63 (0.45 to 0.89) significant (37% reduction) and of
similar magnitude to the effect of other
ref. (188) Mean difference interventions.
4.9 points (2.97 to 6.92)
Intelligence test scores The effect of breastfeeding is
significant, with a substantial effect
size.

CONCLUSIONS

41

References

1. WHO Collaborative Study Team on the 10. Lucas A, Fewtrell MS, Cole TJ. Fetal ori-

Role of Breastfeeding on the Prevention gins of adult disease - the hypothesis re-

of Infant Mortality. Effect of breastfeed- visited. British Medical Journal, 1999,

ing on infant and child mortality due to 319(7204):245-9.

infectious diseases in less developed 11. Lucas A. Programming by early nutrition
countries: a pooled analysis. Lancet, 2000, in man. Ciba Foundation Symposium, 1991,
355(9202):451-5. 156:38-55.

2. Kramer MS, Kakuma R. The optimal du- 12. Gluckman PD, Hanson MA. Living with
ration of exclusive breastfeeding: a sys-
the past: evolution, development, and
tematic review. Advances in Experimental
patterns of disease. Science, 2004,
Medicine and Biology, 2004, 554:63-77.
305(5691):1733-6.

3. World Health Organization. Global Strat- 13. Chalmers I. Unbiased, relevant, and reli-
egy for Infant and Young Child Feeding, The
able assessments in health care: impor-
Optimal Duration of Exclusive Breastfeeding.
tant progress during the past century, but
Geneva, World Health Organization,
plenty of scope for doing better. British
2001.
Medical Journal, 1998, 317(7167):1167-8.

4. World Health Organization and UNICEF. 14. Altman DG et al. The revised CONSORT
Protecting, Promoting and Supporting Breast-
statement for reporting randomized tri-
feeding: The Special Role of Maternity Services.
als: explanation and elaboration. Annals
Geneva, World Health Organization,
of Internal Medicine, 2001, 134(8):663-94.
1989.

5. Barker DJP. Fetal and infant origins of adult 15. Lucas A et al. Multicentre trial on feed-

disease. London, BMJ Publishing, 1992. ing low birthweight infants: effects of diet

on early growth. Archives of Disease in Child-

6. Rich-Edwards JW et al. Birth weight and hood, 1984, 59(8):722-30.

risk of cardiovascular disease in a cohort 16. Kramer MS et al. Promotion of
of women followed up since 1976. British Breastfeeding Inter vention Trial
Medical Journal, 1997, 315(7105):396-400. (PROBIT): a randomized trial in the

7. Newsome CA et al. Is birth weight re- Republic of Belarus. The Journal of the

lated to later glucose and insulin metabo- American Medical Association, 2001,

lism? A systematic review. Diabetic Medi- 285(4):413-20.

cine, 2003, 20(5):339-48. 17. Concato J, Shah N, Horwitz RI.

8. Horta BL et al. Early and late growth and Randomized, controlled trials, observa-

blood pressure in adolescence. Journal of tional studies, and the hierarchy of re-

Epidemiology and Community Health, 2003, search designs. New England Journal of

57(3):226-30. Medicine, 2000, 342(25):1887-92.

9. Joseph KS, Kramer MS. Review of the 18. Victora CG et al. The Pelotas birth co-

evidence on fetal and early childhood hort study, Rio Grande do Sul, Brazil,

antecedents of adult chronic disease. Epi- 1982-2001. Cadernos de Saude Publica,

demiologic Reviews, 1996, 18(2):158-74. 2003, 19(5):1241-56.

42 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES

19. Huttly SR et al. Do mothers overesti- 29. Kark JD et al. Validity of maternal report-
mate breast feeding duration? An exam- ing of breast feeding history and the asso-
ple of recall bias from a study in south- ciation with blood lipids in 17 year olds in
ern Brazil. American Journal of Epidemiology, Jerusalem. Journal of Epidemiology and Com-
1990, 132(3):572-5. munity Health, 1984, 38(3):218-25.

20. World Health Organization. Contemporary 30. Marmot MG et al. Effect of breast-feed-
patterns of breastfeeding. Report on the WHO ing on plasma cholesterol and weight in
Collaborative Study on Breastfeeding. Geneva, young adults. Journal of Epidemiology and
World Health Organization, 1981. Community Health, 1980, 34(3):164-7.

21. Bauchner H, Leventhal JM, Shapiro ED. 31. Kuh D et al. Life course epidemiology.
Studies of breastfeeding and infections. Journal of Epidemiology and Community
How good is the evidence? The Journal of Health, 2003, 57(10):778-83.
the American Medical Association,
1986,256(7):887-92. 32. Grummer-Strawn LM, Mei Z. Does
breastfeeding protect against pediatric
22. Horta BL et al. [Breastfeeding and feed- overweight? Analysis of longitudinal data
ing patterns in two cohorts of children from the Centers for Disease Control and
in southern Brazil: trends and differ- Prevention Pediatric Nutrition Surveillance
ences]. Cadernos de Saude Publica, 1996, 12 System. Pediatrics, 2004, 113(2):e81-6.
(Supplement 1):43-48.
33. Egger M, Smith GD. Bias in location and
23. Victora CG et al. The role of conceptual selection of studies. British Medical Jour-
frameworks in epidemiological analysis: nal, 1998, 316(7124):61-6.
a hierarchical approach. International Jour-
nal of Epidemiology, 1997, 26(1):224-7. 34. Greenland S. Quantitative methods in the
review of epidemiologic literature. Epide-
24. Thompson SG. Why sources of hetero- miologic Reviews, 1987, 9:1-30.
geneity in meta-analysis should be inves-
tigated. British Medical Journal, 1994, 35. DerSimonian R, Laird N. Meta-analysis
309(6965):1351-5. in clinical trials. Controlled Clinical Trials,
1986, 7(3):177-88.
25. Barr SI et al. Effects of increased con-
sumption of fluid milk on energy and 36. Normand SL. Meta-analysis: formulating,
nutrient intake, body weight, and cardio- evaluating, combining, and reporting.
vascular risk factors in healthy older Statistics in Medicine, 1999, 18(3):321-59.
adults. Journal of the American Dietetic Asso-
ciation, 2000, 100(7):810-7. 37. Egger M et al. Bias in meta-analysis de-
tected by a simple, graphical test. British
26. Fomon S. Infant feeding in the 20th cen- Medical Journal, 1997, 315(7109):629-34.
tury: formula and beikost. The Journal of
Nutrition, 2001, 131(2):409S-20S. 38. Begg CB, Mazumdar M. Operating char-
acteristics of a rank correlation test for
27. Promislow JH, Gladen BC, Sandler DP. publication bias. Biometrics, 1994,
Maternal recall of breastfeeding duration 50(4):1088-101.
by elderly women. American Journal of Epi-
demiology, 2005, 161(3):289-96. 39. Sterne JA, Gavaghan D, Egger M. Publi-
cation and related bias in meta-analysis:
28. Eaton-Evans J, Dugdale AE. Recall by power of statistical tests and prevalence
mothers of the birth weights and feed- in the literature. Journal of Clinical Epide-
ing of their children. Human Nutrition miology, 2000, 53(11):1119-29.
Applied Nutrition, 1986, 40(3):171-5.
40. Berkey CS et al. A random-effects regres-
sion model for meta-analysis. Statistics in
Medicine, 1995, 14(4):395-411.

REFERENCES 43

41. Greenland S. Invited commentary: a criti- 51. Morris MC, Sacks F, Rosner B. Does fish
cal look at some popular meta-analytic oil lower blood pressure? A meta-analy-
methods. American Journal of Epidemiology, sis of controlled trials. Circulation, 1993,
1994, 140(3):290-6. 88(2):523-33.

42. Lawes CM et al. Blood pressure and car- 52. Forsyth JS et al. Long chain polyunsatu-
diovascular disease in the Asia Pacific rated fatty acid supplementation in in-
region. Journal of Hypertension, 2003, fant formula and blood pressure in later
21(4):707-16. childhood: follow up of a randomised
controlled trial. British Medical Journal,
43. Boulanger JM, Hill MD. Hypertension and 2003, 326(7396):953.
stroke: 2005 Canadian Hypertension
Educational Program recommendations. 53. Perry IJ, Whincup PH, Shaper AG. Envi-
The Canadian Journal of Neurological Sciences, ronmental factors in the development of
2005, 32(4):403-8. essential hypertension. British Medical
Bulletin, 1994, 50(2):246-59.
44. Forsen T et al. Growth in utero and dur-
ing childhood among women who develop 54. Owen CG et al. Effect of breast feeding in
coronary heart disease: longitudinal study. infancy on blood pressure in later life: sys-
British Medical Journal, 1999, tematic review and meta-analysis. British
319(7222):1403-7. Medical Journal, 2003, 327(7425):1189-95.

45. Brunner EJ et al. Dietary advice for re- 55. Martin RM, Gunnell D, Smith GD.
ducing cardiovascular risk. Cochrane Da- Breastfeeding in infancy and blood pres-
tabase of Systematic Reviews, 2005(4): sure in later life: systematic review and
CD002128. meta-analysis. American Journal of Epidemi-
ology, 2005, 161(1):15-26.
46. Whitten CF, Stewart RA. The effect of
dietary sodium in infancy on blood pres- 56. Martin RM et al. Does breastfeeding in
sure and related factors. Studies of in- infancy lower blood pressure in child-
fants fed salted and unsalted diets for five hood? The Avon Longitudinal Study of
months at eight months and eight years Parents and Children (ALSPAC). Circula-
of age. Acta Paediatrica Scandinavica Supple- tion, 2004, 109(10):1259-66.
ment, 1980, 279:1-17.
57. Martin RM et al. Breastfeeding and car-
47. Singhal A, Cole TJ, Lucas A. Early nutri- diovascular disease risk factors, incidence,
tion in preterm infants and later blood and mortality: the Caerphilly study. Jour-
pressure: two cohorts after randomised nal of Epidemiology and Community Health,
trials. Lancet, 2001, 357(9254):413-9. 2005, 59(2):121-9.

48. Geleijnse JM et al. Long-term effects of 58. Martin RM et al. Breastfeeding and
neonatal sodium restriction on blood atherosclerosis: intima-media thickness
pressure. Hypertension, 1997, 29(4):913-7. and plaques at 65-year follow-up of the
Boyd Orr cohort. Arteriosclerosis, Throm-
49. Koletzko B et al. Long chain polyunsatu- bosis, and Vascular Biology, 2005,
rated fatty acids (LC-PUFA) and perina- 25(7):1482-8.
tal development. Acta Paediatrica, 2001,
90(4):460-4. 59. Lawlor DA et al. Associations of paren-
tal, birth, and early life characteristics
50. Engler MM et al. The effects of a diet with systolic blood pressure at 5 years of
rich in docosahexaenoic acid on organ age: findings from the Mater-University
and vascular fatty acid composition in study of pregnancy and its outcomes.
spontaneously hypertensive rats. Circulation, 2004, 110(16):2417-23.
Prostaglandins, Leukotrienes, and Essential Fatty
Acids, 1999, 61(5):289-95.

44 EVIDENCE ON THE LONG-TERM EFFECTS OF BREASTFEEDING: SYSTEMATIC REVIEWS AND META-ANALYSES